Contents
Lock-Free Parallel Access Collections
Bruce P. Lester,
1-12
13-28
29-38
39-50
51-57
59-67
69-76
77-85
87- 96
ISSN: 9081-2546
Journal of Distributed and Parallel Systems , Vol. 7, Issue 2, 2014
ABSTRACT
All new computers have multicore processors. To exploit this hardware parallelism for improved
performance, the predominant approach today is multithreading using shared variables and locks. This
approach has potential data races that can create a nondeterministic program. This paper presents a
promising new approach to parallel programming that is both lock-free and deterministic. The standard
forall primitive for parallel execution of for-loop iterations is extended into a more highly structured
primitive called a Parallel Operation (POP). Each parallel process created by a POP may read shared
variables (or shared collections) freely. Shared collections modified by a POP must be selected from a
special set of predefined Parallel Access Collections (PAC). Each PAC has several Write Modes that
govern parallel updates in a deterministic way. This paper presents an overview of a Prototype Library
that implements this POP-PAC approach for the C++ language, including performance results for two
benchmark parallel programs.
KEYWORDS
Parallel Programming, Lock-free Programming, Parallel Data Structures, Multithreading.
1. INTRODUCTION
The National Research Council of the USA National Academies of Sciences recently issued a
report entitled The Future of Computing Performance [1]. Following is a brief excerpt from the
Preface of this report (p. vii):
Fast, inexpensive computers are now essential for nearly all human endeavors and have been a
critical factor in increasing economic productivity, enabling new defense systems, and advancing
the frontiers of science. ... For the last half-century, computers have been doubling in
performance and capacity every couple of years. This remarkable, continuous, exponential
growth in computing performance has resulted in an increase by a factor of over 100 per decade
and more than a million in the last 40 years. ... The essential engine that made that exponential
growth possible is now in considerable danger.
The response of the computer processor manufacturers (such as Intel) to this challenge has been
the introduction of chip multiprocessors with multiple processing cores per chip, usually called
multicore processors. Each core is essentially a miniature processor, capable of executing its own
independent sequence of instructions in parallel with the other cores. Processors for new laptop
computers now typically have 2 to 4 cores, processors for desktops have 4 to 6 cores, and highperformance processors have 8 to 12 cores. The number of cores per processor is expected to
double every two to three years. This gradually increasing hardware parallelism offers the
potential for greatly improved computer performance. Therefore, the NRC Report concludes the
following (p. 105):
1
Future growth in computing performance will have to come from software parallelism that can
exploit hardware parallelism. Programs will need to be expressed by dividing work into multiple
computations that execute on separate processors that communicate infrequently or, better yet,
not at all.
To achieve this goal, the NRC Report gives the following recommendation:
Invest in research in and development of programming methods that will enable efficient use of
parallel systems not only by parallel systems experts but also by typical programmers. (p. 138)
The purpose of this paper is to address this critical need in computing today, as expressed in the
NRC Report. We explore a new high-level parallel programming abstraction that is easy to use
and achieves good performance, without the need for locking and without the possibility of data
races that can cause nondeterministic program execution.
2. B ACKGROUND
The predominant approach to parallel programming in current computer technology is
Multithreading: the programming language has special primitives for creating parallel threads of
activity, which then interact through shared variables and shared data structures. To prevent
conflicts in the use of the shared data, locking is used to provide atomic access. Examples of
popular multithreading systems are OpenMP [2] and Java Threads [3]. One of the main problems
with this multithreading approach is data races that result when parallel threads update the same
memory location (or one thread reads and the other updates). Depending on which thread wins
the race, the final result may be different. This essentially creates a nondeterministic program: a
program that may produce different outputs for the same input data during different executions.
This nondeterminism complicates the software development process, and makes it more difficult
to develop reliable software [4]. This introduction of data races and nondeterminism into
mainstream computer programs is hardly a step of progress in computer programming, and is
considered by many to be unacceptable (see S. Adve, Data Races are Evil with No Exceptions
[5]).
In multithreaded parallel programming, locking is generally used to control access to shared data
by parallel threads. If correctly used, locking helps to maintain the consistency of shared data
structures. However, locking introduces other problems that complicate parallel programming.
Even if used correctly, multiple locks used to protect different data structures can interact in
unfortunate ways to create a program deadlock: a circular wait among a group of parallel threads
which ends program execution prematurely. Furthermore, these deadlocks may occur in a
nondeterministic manner, appearing and then disappearing during different executions of the
same program with the same input data. Of course, there is a wide variety of proposed solutions
to this deadlock problem, but none is completely effective or has achieved wide acceptance in
practice [6].
Another problem with locking is performance degradation. The locking operation itself takes
considerable time and is an additional execution overhead for the parallel program. Also, the lock
by its very nature restricts parallelism by making threads wait, and thus reduces parallel program
performance and limits the scalability of the program. (A parallel program is said to be scalable if
the execution time goes down proportionately as more processors, or cores, are added.) There has
been some research on lock-free parallel access data structures [7], but with limited success so
far.
We have developed a new high-level parallel programming abstraction that does not require
locking and does not allow data races. With this new abstraction, parallel programs are
guaranteed to be deterministic and deadlock-free even in the presence of program bugs. In the
following sections, we will describe the details of this new approach to parallel programming.
3. O UR A PPROACH
Anyone who has been a serious parallel programmer or parallel algorithm designer knows that the
best source of scalable parallelism is program loops. Computationally intensive programs, which
are candidates for parallel execution, typically have program loops (or nested loops) in which the
loop body is executed many times sequentially. If the loop iterations are fairly independent, they
can be executed in parallel by different processor cores to significantly speedup program
execution. This property of programs was recognized early in the history of parallel computing.
Early parallelizing compilers, such as Parafrase [8], had some limited success with automatic
parallelization of sequential program loops. Numerical programs that operate on large multidimensional data arrays typically have lots of such loops that in principle can be parallelized.
Unfortunately, the automatic parallelization by compilers has shown itself to be too limited. More
often than not, a programmer is needed to restructure the underlying algorithm to expose a more
complete level of parallelism [9].
This has led to explicit parallel loop constructs in many programming languages (or APIs for
parallel programming). Recent versions the Fortran language have a DOALL directive for parallel
DO-loop execution [10]. Many parallel versions of the C programming language have some kind
of forall instruction, which is a parallel form of the ordinary sequential for-loop [11]. The popular
OpenMP standard for parallel programming in the C language [2] has a parallel-for directive for
assigning sequential for-loop iterations to different threads for parallel execution. Unfortunately,
these parallel loop instructions alone do not solve the problem of data races for access to shared
data structures. Therefore, additional primitives such as locks are needed to synchronize the
parallel processes. This introduces the problems of deadlock and performance degradation as
described earlier. Also, if the process synchronization is not done completely correctly by the
programmer, data races can remain in the parallel program causing nondeterministic execution.
We feel the parallel for-loop is a good step in the right direction moving from sequential to
parallel programming, but does not go far enough. A more highly structured (or more abstract)
form of the parallel-for is needed to prevent data races and guarantee deterministic execution
without explicit locking by the programmer. We have developed a new high-level parallel
programming abstraction for the C/C++ language that has an operational component and a data
component. The operational component is called a Parallel Operation (abbreviated POP). The
data component is called a Parallel Access Collection (abbreviated PAC).
An ordinary program loop has three basic categories of data:
Local Data: variables (or data structures) declared local to each loop iteration and only
accessible by one loop iteration.
Read-only Shared Data: variables (or data structures) read by multiple loop iterations,
but not updated by any loop iterations.
Write Shared Data: variables (or data structures) accessed by multiple loop iterations and
updated by at least one loop iteration.
Local data and Read-only shared data do not pose a problem for parallel loop execution, and thus
can be accessed freely within a POP. Data races and consequent nondeterminacy arise from Write
Shared Data. Therefore, access to Write Shared Data must be carefully controlled. In our new
parallel programming abstraction, all Write Shared Data of any POP must be selected from a
special category of data structures called Parallel Access Collections (PACs). The POPs and
PACs work together to allow deterministic access to the Write Shared Data of the parallel loop
iterations.
3
4. POP-PAC LIBRARY
We have implemented a prototype library for this POP abstraction in the C++ programming
language with three types of PACs. This library was then used for performance testing of two
parallel benchmark programs. In the following sections, we will describe the details of this library
and the benchmarks.
The three types of PACs implemented in the C++ library are as follows:
pArray: A simple one-dimensional array where each element has the same base type
pArray2: A standard two-dimensional version of the pArray
pList: An extension of pArray where elements can be appended to the end of the list
Each of these types of PACs is capable of functioning in either of the two Write Modes (Private
or Reduce). Each PAC is represented in the C++ library as a template class. The prototype library
also contains several template classes to implement the POP operation. We used this prototype
library to implement two parallel algorithms:
Shortest Path in a Graph: This parallel algorithm uses three pArray PACs all in Reduce
Mode.
Jacobi Relaxation to solve a Partial Differential Equation: This parallel algorithm
uses a total of three pArray2 PACs. Two of the PACs are in Private Mode, and one in
Reduce Mode.
We will begin our discussion of the feasibility study by describing the details of the parallel
program for determining the Shortest Path in a weighted, directed graph. To simplify the
algorithm, only the shortest distance from the source vertex 0 to every other vertex will be
computed, and stored in a one-dimensional array mindist. To represent a graph with n vertices,
use an n by n two-dimensional array called weight. The value of weight[x][w] will give the weight
of the edge from vertex x to vertex w in the graph. The algorithm for finding the shortest distances
is based on gradually finding new shorter distances to each vertex. Whenever a new short
distance is found to a vertex x, then each neighboring vertex w of vertex x is examined to
determine if mindist[x] + weight[x][w] is less than mindist[w]. If so, a new shorter distance has
been found to vertex w via x. Vertex w is then put into a queue of vertices that have to be further
explored. When the neighbors of x have all been considered in this way, the algorithm selects a
new vertex x from the queue and examines all of its neighbors. When the queue is empty, the
algorithm terminates and the mindist array contains the final shortest distance from the source
vertex to every other vertex. A sequential version of this shortest path algorithm is shown in
Figure 1 (ref. [11], p. 412).
For a large graph, the queue will grow in size very quickly and then gradually diminish as the
algorithm progresses until it is eventually empty and the algorithm terminates. The sequential
5
void main( ) {
initialize mindist array to infinity;
initialize inflag array to False;
inflag[0] = True; mindist[0] = 0; // set vertex 0 as source
Set Write Mode of mindist, inflag, outflag, more to Reduce;
Set Reduction Function of mindist to minimum( );
Set Reduction Function of inflag, outflag, more to logicalOR( );
do {
more = False;
POP ( n, ShortPathFunction, &mindist, &outflag, &more );
Exchange outflag and inflag arrays;
} while (!more)
}
// this function forms the body of each process
void ShortPathFunction( int x, mindist, outflag, more ){
if (inflag[x]) { // if vertex x is in the queue
foreach neighboring vertex w of x {
newdist = mindist[x]+ weight[x][w];
if (newdist < mindist[w]) {
// found a new shorter distance to w via x
mindist[w] = newdist;
outflag[w] = True;
more = True;
}
}
inflag[x] = False; // vertex x is removed from queue
}
Figure 2 Parallel Shortest Path Algorithm
algorithm is naturally parallelized by considering all the vertices in the queue in parallel. Let us
use an inflag[n] array of True and False values, which indicates whether each vertex of the graph
is currently in the queue. Then a POP can be created that assigns one process to each element of
the inflag array. If any element inflag[x] is True, then the process examines the neighboring
vertices w of vertex x, as described above for the sequential algorithm. If a new shorter path is
found to vertex w, then mindist[w] is updated, vertex w is added to the queue, and outflag[w] is
set to True. When the current POP finishes, then the outflag array replaces the inflag array and
becomes the starting point for the next POP. These POPs are iteratively repeated until an outflag
array of all False results.
In an ordinary parallel implementation of this algorithm using multithreading, locking would be
needed to prevent conflicts during updates of the arrays. However, by using PACs with the proper
Write Modes, locking is not needed in our parallel formulation. This parallel shortest path
algorithm requires three PACs: inflag, outflag, and mindist are each represented as an integer
pArray with n elements. To simplify the termination detection, an additional integer pArray called
more is used with one element. All of the pArrays are set to Reduce Mode. Since the inflag array
indicates which vertices are currently in the queue, a queue data structure is not actually needed in
the parallel algorithm. A centralized queue would pose a performance bottleneck and limit
scalability. A high-level pseudo-code description of the parallel algorithm is shown in Figure 2.
The main( ) function contains the initializations and the high-level do-while loop that drives the
algorithm. Each loop iteration contains one POP that creates n parallel processes: one for each
Number of Cores
vertex of the graph. Each process is given its own unique index from 0 to n-1. The
ShortPathFunction( ) forms the body of each process. Notice that the ShortPathFunction( ) is
almost identical to the sequential version of the Shortest Path Algorithm shown in Figure 1. The
inflag array is a PAC, but it is read-only in ShortPathFunction( ), and thus does not have to be
included in the list of input PACs. It is possible that two parallel processes may update the same
element in the mindist array. This potential for a data race is resolved by the minimum( ) function,
which is the Reduction Function assigned to the mindist array. Since minimum( ) is associative
and commutative, the final result in mindist will be independent of the relative timing of the
process updates to mindist. The same is true of the PACs outflag and more, which have
logicalOR( ) as the Reduction Function. Thus, no locking or other process synchronization is
needed in the parallel algorithm. All parallel writes to the PACs are resolved in a deterministic
way by the assigned Write Modes.
As part of our feasibility study, we implemented a prototype C++ library for the POP and the
three types of PACs described above (pArray, pArray2, pList). The POP and PACs are all
implemented as template classes in the library. The library also contains some additional helper
template classes to implement the Write Modes. This POP and PAC library was used to execute
the Parallel Shortest Path Algorithm shown in Figure 2 on a Dell Studio XPS workstation with a
six-core AMD Opteron processor. Figure 3 shows the performance results for a graph with 4000
vertices, as the number of cores applied to the parallel computation is varied from one to six. The
graph in Figure 3 shows that the POP program for Shortest Path does appear to scale well: the
program runs 4.8 times faster using six cores than using one core.
2 v 2v
0
x 2 y 2
This equation can be solved numerically using a two-dimensional array of discrete points across
the surface of the metal sheet. Initially, the points along the boundaries are assigned the
appropriate constant voltage. The internal points are all set to 0 initially. Then Jacobi Relaxation
is used to iteratively recompute the voltage at each internal point as the average of the four
void main( ) {
initialize pArray2 A;
Set Write Mode of A, B to Private;
Set Write Mode of pArray maxchange to Reduce;
Set Reduction Function of maxchange to maximum( );
tolerance = .00001;
do {
// perform 10 Jacobi iterations
for ( i = 0; i < 10; i++) {
POP ( n, JacobiFunction, &B );
Exchange A and B arrays;
}
// convergence test every ten iterations
POP ( n, ConvergenceTestFunction, &maxchange);
}
immediate neighboring points (above, below, left, right). Convergence is tested by comparing a
desired tolerance value to the maximum change in voltage across the entire grid during each
iteration.
The parallel version of this program requires two different POP operations: one for performing
the Jacobi iterations, and another for the Convergence test. Three different PACs are required:
Array A[n][n] contains the current value at each point at the start of the iteration
Array B[n][n] holds the newly computed value at each point (average of four neighboring
points)
Scalar maxchange holds the maximum change across the whole array during a given
iteration
8
Number of Cores
Arrays A and B are represented in the parallel program as pArray2 PACs in Private Mode.
Variable maxchange can be represented as a pArray (with one element) in Reduce Mode. The
parallel program is shown in Figure 4. The main( ) function drives the whole program with two
nested loops. The first loop performs ten Jacobi iterations using a POP that calls the
JacobiFunction( ). The ten iterations are sequential, but within each iteration, a separate parallel
process is assigned to compute each point of the two-dimensional grid. Since only one parallel
process writes to each point of pArray2 B, Private mode is used for the Write Mode. After ten
Jacobi iterations, the second POP is invoked to perform the convergence test by subtracting the
values in the corresponding locations of A and B. By having each process write its change directly
into maxchange, the overall maximum value of the change is computed because maxchange is set
to Reduce Mode with reduction function maximum( ).
The Jacobi Relaxation program of Figure 4 was tested using an array of 40,000 points on a Dell
Workstation with a six-core processor. The graph of Figure 5 shows the performance results as
the number of cores assigned to the program is varied from one to six. The results are similar to
the Shortest Path program. The parallel Jacobi program appears to exhibit good scalability: the
program runs 4.4 times faster using six cores than one core.
6. R ESEARCH COMPARISON
There has been a considerable amount of research over the years in lock-free parallel
programming and deterministic parallel programming. We will briefly focus on some of the
major trends in this research that are most relevant to our proposed research direction. One
important approach to lock-free parallel programming is Software Transactional Memory (STM)
[ref. 12, 13, 14]. With STM, the programmer can specify certain portions of code to be
transactions. The STM implementation will automatically guarantee that each transaction is
executed in an atomic way, without interference from other transactions. This is usually
implemented with some form of optimistic scheduling: parallel transactions may proceed freely
until some possibility of conflict is detected. Then one of the conflicting transactions is rolled
back and rescheduled. This guarantees a serializable execution of parallel transactions
equivalent to some serial (sequential) execution of the transactions. The main problem with STM
has been the implementation overhead that slows parallel program execution.
In our opinion, Software Transactional Memory has some usefulness, but does not go far enough.
A serializable schedule of transactions is not necessarily deterministic. For example, consider two
parallel transactions to update a variable x with initial value 50: transaction A reads the current
value of x, squares it, and writes the result back to x; transaction B reads the current value of x,
adds 10, and writes the result back to x. If Transaction A executes first then B, the result is x =
2510. If Transaction B executes first then A, the result is x = 3600. Both schedules are
serializable, and thus permitted by STM. Since both transactions read and write variable x, there
is a possibility of interference between the transactions. This kind of interference is prevented by
STM, but it still does not guarantee determinacy. STM eliminates some of the data races in a
parallel program, but not all. Thus, the parallel programmer will still be faced with debugging
nondeterministic programs that may produce different outputs using the same input data.
Our proposed use of POPs and PACs for parallel programming is really quite different from
Software Transactional Memory. The POP is a very specific way of creating a team of parallel
process. The PACs are special data structures with Write Modes. Whereas, with STM, the
programmer just inserts some begin and end transaction primitives into the program. These can be
inserted anywhere and apply to any type of data. Our PACs have Reduce and Private Mode
there is nothing like this in Software Transactional Memory.
Another notable trend of research in parallel programming that has some relevance to our
research on POPs and PACs is data parallel programming [15, 16, 17], sometimes called
Collection-Oriented (CO) parallel programming [18]. With CO programming, certain
commonly used highly parallel operations on collections (data structures) are built into a
collections library. This saves the parallel programmer from having to reinvent the wheel in
every program. Some simple examples of such highly parallel operations are Broadcast, Reduce,
Permute, Map, Scatter, Gather [19]. These standard operations are implemented internally in the
collections library in a highly parallel way. As is the case with ordinary sequential program
libraries, this approach is very useful and improves programmer productivity. However, practical
experience has shown that the standard library operations alone are not sufficient to structure a
whole parallel program. Often the parallel program requires some custom operations that are
not easily created from a combination of the available collections library operations. Thus, our
proposed POPs and PACs are not a replacement for collection-oriented parallel programming, but
a supplement to it. The two approaches are complementary and can be used together in the same
program.
Another research direction for deterministic parallel programming is deterministic scheduling of
multithreaded programs with shared variables [20, 21, 22, 23]. This approach allows the data
races to remain in the parallel program, but forces the data races to be resolved in a repeatable
(deterministic) manner. One major drawback of this approach is the resultant parallel programs
are fragile: even an insignificant modification in the program code can change the outcome of a
data race in some unrelated portion of the program, thus changing the final outcome of the
program. As long as the program is not modified at all, the deterministic scheduling will
guarantee a repeatable outcome for all the data races, but a small change in the code may expose a
data race bug in a completely different portion of the program. Our POP-PAC approach is quite
different than this deterministic scheduling approach. Our approach is more highly structured
because we have a specific parallel control structure (POP) and specific parallel access data
structures (PACs) with many Write Modes.
7. F UTURE RESEARCH
As defined in Section 3, each POP has an associated function, which is the executable code
supplied by the User. Local variables are (non-static) variables declared inside the function body.
Variables defined outside the function body are considered as shared variables. For the POP to be
deterministic, the User's POP function must satisfy the following simple rules:
10
These are fairly simple rules for a programmer to follow in order to guarantee determinacy of
each POP. However, the prototype library we describe in this paper does not enforce these rules.
Therefore, it is possible for bugs in the User function to violate these rules and cause
nondeterministic execution of the POP. The next phase of our research on the POP-PAC approach
will be to modify the C++ compiler (and language runtime system) to enforce these rules. Then
all program POPs will be guaranteed to be deterministic even in User programs with bugs.
REFERENCES
[1]
[2]
[3]
[4]
[5]
[6]
[7]
[8]
[9]
[10]
[11]
[12]
[13]
[14]
[15]
[16]
[17]
[18]
[19]
[20]
[21]
S. H. Fuller and L. I. Millett, Editors, (2011) The Future of Computing Performance: Game Over or
Next Level?, National Research Council, National Academies Press.
B. Chapman, et al., (2007) Using OpenMP: Portable Shared Memory Parallel Programming. MIT
Press.
C. Horstmann and G. Cornell, (2007) Core Java, Volume I--Fundamentals (8th Edition). Prentice
Hall, pp. 715-808.
E. Lee, The problem with threads,(2006) Computer, Vol. 39, No. 5, pp. 3342.
S. Adve, (2010) Data Races are Evil with No Exceptions, Communications of the ACM, vol. 53, no.
11, p. 84.
S. Bensalem, et al., (2006) Dynamic Deadlock Analysis of Multi-threaded Programs, In
Proceedings First Haifa International Conference on Hardware and Software Verification and
Testing, pp. 208-223.
N.Shavit, (2011) Data Structures in the Multicore Age, Communications of the ACM, vol. 54, no.
3, p. 76-84.
Polychronopoulos, C., et al., (1989) Parafrase-2: An Environment for Parallelizing, Partitioning,
Synchronizing and Scheduling Programs on Multiprocessors, In Proceedings 1989 International
Conference on Parallel Processing, Vol. II, pp. 39-48.
G. Tournavitis, et al., (2009) Towards a Holistic Approach to Auto-Parallelization presented at
ACM SIGPLAN Conference on Programming Language Design and Implementation (PLDI 2009),
Toronto, Canada, pp. 177-187.
Fortran Programming Guide, (2001) Sun Microsystems, Palo Alto, CA.
B. Lester, (2006) The Art of Parallel Programming (Second Edition). First World Publishing.
A. Dragojevic et al., (2011) Why STM Can Be More Than a Research Toy, Communications of the
ACM, vol. 54, no. 4, pp. 70-77.
J. Larus and C. Kozyrakis, (2008) Transactional Memory, Communications of the ACM, vol. 51, no.
7, pp. 80-88.
C. Cacaval, et al. , (2008) Software Transactional Memory: Why is it Only a Research Toy?
Communications of the ACM, vol. 51, no. 1, pp. 40-46.
G. E. Blelloch, (1990) Vector Models for Data-Parallel Computing. The MIT Press, Cambridge,
Massachusetts.
G. Tanase, et. al., (2011) The STAPL Parallel Container Framework, In Proceeding ACM
SIGPLAN Symposium on Principles and Practice of Parallel Programming (PPOPP), pp. 235-246.
Array Building Blocks Application Programming Interface Reference Manual. (2011) Intel
Corporation.
B. Lester, (2011) Improving Performance of Collection-Oriented Operations through Parallel
Fusion, In Proceedings of The World Congress on Engineering 2011, pp. 1519-1529.
B. Lester, (1993) The Art of Parallel Programming (First Edition). Prentice Hall.
T. Bergan et al. (2010) CoreDet: A compiler and runtime system for deterministic multithreaded
execution, In Proceedings Architectural Support for Programming Languages and Operating
Systems (ASPLOS 2010), pp. 53-64.
M. Olszewski, J. Ansel, and S. Amarasinghe, (2009) Kendo: Efficient Deterministic Multithreading
in Software, In Proceedings Architectural Support for Programming Languages and Operating
Systems (ASPLOS 2009), pp. 97-108.
11
[22] E. D. Berger et al., (2009) Grace: Safe multithreaded programming for C/C++, In Proceedings
ACM SIGPLAN Conference on Object Oriented Programming Systems Languages and Applications
(OOPSLA 2009), pp. 81-96.
[23] H. Cui, J. Wu, and J. Yang. (2010) Stable deterministic multithreading through schedule
memoization, In Proceedings 9th USENIX Conference on Operating Systems Design and
Implementation (OSDI 10).
Author
Dr. Bruce Lester received his Ph.D. in Computer Science from M.I.T. in 1974. He was a Lecturer in the
Department of Electrical Engineering and Computer Science at Princeton University for two years. Dr.
Lester is one of the pioneering researchers in the field of parallel computing, and has published a textbook
and numerous research papers in this area. He founded the Computer Science Department at Maharishi
University of Management (MUM), where he has been a faculty member for sixteen years.
12
ISSN: 9081-2546
Journal of Distributed and Parallel Systems , Vol. 7, Issue 2, 2014
ABSTRACT
In this paper we analyze the methodology developed by the International Telecommunication Union (ITU)
for estimating the spectrum requirement for International Mobile Telecommunications (IMT) systems. The
International Telecommunication Union estimates spectrum requirements by following ITU-R-Rec.M1768.
Although this methodology is adopted by ITU-R, there are discrepancies for estimating the spectrum
requirement for developing countries. ITU estimates the spectrum requirement by considering technical
and market parameters that were provided by the most developed countries with high income and high
development index. Developed countries have a very rapid expansible telecom market due to the high level
of penetration, dominant user density and usage of high-volume multimedia services. In contrast,
developing countries use less bandwidth-intensive services such as voice communication, low rate data,
low and medium multimedia. However, while the input parameters are adequate for developed countries,
they do not reflect the status of developing countries. For this reason the ITU spectrum estimation
overestimates the exact requirements of spectrum for IMT systems for developing countries. This paper
presents an approach based on the technical and market related parameters, which is thought to be
applicable for overcoming the shortcomings of the current ITU methodology in estimating the spectrum
requirement for developing countries like Bangladesh.
KEYWORDS
International Mobile Telecommunications (IMT), Radio Access Technique Groups (RATGs), Spectrum
Estimation, Spectrum Management.
1. INTRODUCTION
To estimate the radio spectrum for IMT Systems, International Telecommunication Union
adopted the methodology described in ITU-R Rec M.1390 [1]. Spectrum requirements for the
terrestrial components of IMT-2000 were estimated in Report ITU-R M.2023 [2] prior to
WRC-2000 by using this methodology. ITU-R.M1390 has been considered to be a framework
focusing on a single system and market scenario based on blended 2G and 3G networks. In
addition, this methodology was essentially based on the paradigm of circuit switching. But this
approach is no longer suitable for meeting the current requirements set by ITU for IMT Systems,
since packet- based traffic now dominates as the switching scheme in current and conventional
wireless networks. The methodology ITU-R Rec M.1390 [1] does not properly consider packetbased traffic. As stated in Rec. ITU-R M.1645 [3] the majority of traffic has been shifting from
voice oriented communications to multimedia communications with Internet Protocol-based
packet switching. To consider the changes in wireless service environment a new methodology,
ITU-R Rec.M1768 has been developed by ITU for planning and calculating spectrum use in IMT
Systems [4]. This new method specifically emphasizes the capacity of a cell of wireless system
that needed to fulfill Quality of Service (QoS) requirements of an arbitrary number of different
13
service classes offering packet traffic. The methodology is applicable to both circuit and packet
switch-based traffic and can accommodate multiple services. The Report ITU-R M.2072 [4]
provides a summary of the market analysis and forecast for the evolution of mobile markets and
services for the future development of IMT systems and derives market related parameters for the
years of 2010, 2015 and 2020. To calculate the spectrum requirements, the report analyzed data
from thirty countries and organizations and provided typical data. Bangladesh was one of the
developing countries analyzed. In 2004, the cellular mobile penetration rate in Bangladesh was
only 3.7%, compared to 68% in Japan and 86% in Republic of Korea in roughly the same time
frame.
It is obvious that spectrum requirements for cellular mobile communications are directly
proportional to the size of the telecommunication market. The market size of IMT systems can be
estimated from many kinds of parameters. For example, the total population is an important
driver of cellular penetration: higher population density lowers the cost of service delivery.
Telecommunication growth is also highly correlated with economic indicators such as the GDP,
GNP and the overall economic development of a country. Income and literacy rate are the key
market parameters for market estimation, since they can be used to approximate the average
citizens ability to buy mobile telecommunication services. The technology available to users at a
certain point of time also influences the market because new technologies lead to new services
and create demand.
ITU-R-Rec.M1768 estimates the spectrum requirements based on technical parameters such as
the cell area, application data rate, radio parameters, population coverage percentage and market
parameters like user density, session arrival rate per user, mean service bit rate, average session
duration, market settings and mobility ratio are considered in Reports ITU-R.M2072, ITUR.M2074 and ITU-R.M2078. These parameters are typically based on the data provided in the
most developed parts of the world where the saturation level of market development may be
reached earlier than the global market and where user expectations are continually increasing for
their consuming capability with regard to the variety of multimedia services and applications.
Due to differences in market development, the input parameters should be interpreted differently
for developing countries at different time intervals. Thus the technical and market parameters
have to be changed to estimate the spectrum requirements for developing countries. This paper
presents an acceptable approach to overcome the shortcomings of the available ITU spectrum
estimation methodology that can be applicable to developing countries in order to avoid
overestimating the requirements. It also proposes a scenario for spectrum planning for IMT
Systems for a developing country like Bangladesh in terms of increasing wireless communication
traffic.
The remainder of this paper is organized as follows. Section 2 gives an overview of the ITU-R
spectrum estimation methodology and its technical and market parameters. Section 3 presents an
approach for estimating the spectrum requirement for developing countries by using relevant
technical and market data for Bangladesh. Section 4 explains the spectrum requirement
estimation for Bangladesh and compares the obtained result with the spectrum requirements
forecasted by ITU. Finally, the proposed method and its results are summarized in section 5.
2. LITERATURE REVIEW
In 2000, ITU estimates 530.3 MHz spectrum to be required for IMT systems (Methodology ITUR Rec M.1390). But this methodology does not properly consider broadband data and multimedia
traffic. To overcome this problem ITU has been developed a new methodology ITU-R
Rec.M1768 through IST-WINNER project (Marja Matinmikko1, Tim Irnich, Jrg Huschke, Antti
Lappetelinen, and Jussi Ojala). This methodology forecasted 1720 MHz spectrum to be required
14
for IMT 2000 and IMT advanced systems. According to Chung, Lim, Yook and Park (2007) who
forecasted spectrum requirement based on the Korean communication environment, they have
shown that in the case of spectral efficiency using higher modulation and coding schemes, the
spectrum requirement of IMT advanced is approximately 2700 MHz. When applying 2x2
multiple-input multiple output (MIMO) antenna system, it is approximately 1500 MHz. When
applying 4x4 multiple-input multiple output (MIMO) antenna system, it is approximately 1050
MHz. In 2013, a new model for spectrum estimation has been commissioned by the GSM
Association to estimate future spectrum requirements for different countries and mobile markets
by taking a theoretical approach, the starting point of the model is the existing number of base
station sites within a given market and scope for future site densification. By this methodology
they estimate spectrum requirement for four countries and the preliminary results indicate a total
spectrum of 1939 MHz will be required for USA, 2074 MHz for UK, 2080 MHz for Brazil and
1844 MHz for China.
15
Conversational
SC1
SC2
SC3
SC4
SC5
Streaming
SC6
SC7
SC8
SC9
SC10
Interactive
SC11
SC12
SC13
SC14
SC15
Background
SC16
SC17
SC18
SC19
SC20
Dense urban
SE1
SE2
SE3
Teledensity
Sub-urban
SE4
SE5
Rural
SE6
Radio environments are characterized by two parameters: cell area and population
coverage percentage. The cell area of a radio environment may vary depending on the
teledensity [7]. A cell is the area covered by a base station. The cells can be classified
according to their size as macro, micro, pico and hotspot cells. The cell area largely
depends on the user density, type of service and transmitter output power and has a direct
impact on the traffic volume and spectrum requirements. It is also closely related to other
radio parameters such as the application data rate and the minimum deployment per
operator per radio environment.
3.1.4. Radio access technique groups
RATG 1: Pre-IMT systems, IMT-2000 and its enhancements. This group covers the digital
cellular mobile systems, IMT-2000 systems and their enhancements.
RATG 2: Systems beyond IMT-2000 (e.g., new mobile access and new nomadic/local area
wireless access), but not including systems already described in any other RATGs [4].
3.1.5 Area spectral efficiency
The spectral efficiency is an important parameter for converting the capacity requirements in
terms of bit/s/cell to the spectrum requirements in Hz. If the spectral efficiency is increased, the
unadjusted and final spectrum requirements decrease and vice versa [7].
16
17
Td ,t,n,rat, p .
(1)
sn
Let C (bit/s) be the capacity of the channel. Then the mean bn (s/packet) and the second
moment of
of the transmission time for an IP packet of service category n are given by bn =
and
= The mean delay of an IP packet of service category n is given by
Nps
Dn (C)
s
i1
(2)
i i
2C i si C i si
i1 i1
n1
sCn
(2)
This equation is derived from Cobhams formula for the mean waiting time in a single arrival
M/G/1 non-pre-emtive priority queue [Cobham et al., 1954] [14], [Kesten and Runneberg et al.
1957] [15], [Inrich and Walke et al., 2004] [16]. If the QoS requirement for IP packets of class
n, is given in terms of a required mean delay Dn, the required system capacity Cn is defined as
the capacity satisfying the condition D(Cn) = Dn. Given a certain value for Dn, the system
capacity Cn required to achieve Dn can be calculated by solving (2) for C. Among the three
roots of this equation there is always one that satisfies the stability condition
n
i si
i 1
(3)
Cn
This value is chosen as the solution to the problem of dimensioning the system capacity so that IP
packets of service class n have the required mean delay Dn. The system capacity that fulfills the
mean delay requirements of all classes, denoted by C, is obtained by determining the set {C1,
C2.CN} of system capacity values required to fulfill the QoS requirements of classes n = 1-N, and selecting the maximum value from this set, i.e. C= max (C1, C2.CN).
(4)
The capacity requirements are now converted to the spectrum requirements by dividing the area
spectral efficiency factors. The spectrum requirement is obtained by
F d ,t , rat , p
C d ,t , rat , p
d , rat , p
(5)
.Fd,t,rat + (No 1)
(6)
where
(Hz) is the guard band between operators. The total spectrum requirements of RATG1
and RATG2 are taken as the maximum over the teledensities (dense urban, suburban, rural) [4].
18
).
and micro cells and all high mobility to macro cells. The respective radio environment should be
available, otherwise the traffic would go to the radio environment with the next higher mobility
support. Some radio environments can be supported in all service environments. In practice, the
total area of a service environment is covered by a certain radio environment, only up to a certain
percentage, which is defined as the population coverage percentage [7]. For the years of 2010,
2015 and 2020, ITU proposed population coverage percentages for different radio environments
in different service environments in Table 4 [7]. This is satisfactory for service environments
included in the dense urban area but not harmonious for service environments in sub-urban and
rural areas of a developing country. They can be covered under a cellular mobile network by
using macro or micro cells because of the absence of high rise buildings and other obstacles. In
addition to macro and micro cells, pico cell can be installed in office and public area where large
capacity is required. According to the ITU methodology, pico cells and hotspots are proposed for
rural areas, but this is not appropriate for developing countries. The base station antenna height of
pico cells and hotspots may be much lower than the average rooftop height, so pico cells and
hotspots are not feasible for rural areas. Network coverage is often lacking in low population
density and low income rural areas of the developing world because of the expense of
infrastructure. To solve this coverage gap, we propose using only macro cells for rural and remote
areas. In some special cases micro cells can be installed in village areas that often feature a single
community center where schools, hospitals and markets are located.
Table 4. ITU proposed population coverage percentage.
Service
Environment
s
SE1
SE2
SE3
SE4
SE5
SE6
Year
Macro Cell
2010
2015
2020
2010
2015
2020
2010
2015
2020
2010
2015
2020
2010
2015
2020
2010
2015
100
100
100
100
100
100
100
100
100
100
100
100
100
100
100
100
100
Radio environments
Micro Cell
Pico Cell
90
90
90
90
90
90
95
95
95
15
35
35
40
50
50
0
0
0
10
20
20
20
20
20
30
40
0
0
0
35
35
35
10
10
Hot Spot
80
80
80
80
80
80
10
25
40
80
80
80
20
20
20
50
50
environments are included in RATG 1 for indoor coverage in dense urban areas. Network
capacity and spectrum requirements largely depend on the application data rate and the mean
service bit rate. The application data rate proposed by ITU for RATG 1 is acceptable but it is not
typical for RATG 2 in developing countries. ITU estimated the application data rate for spectrum
as 50 Mbit/s for macro cell, 100 Mbit/s for micro cell and 1 Gbit/s for pico cell/hotspot for RATG
2 [5]. Data rates of 30 Mbit/s to 100 Mbit/s or 1Gbit/s are required only for super high multimedia
services. However overall multimedia services such as high quality video conferencing,
multimedia phone, mobile HDTV, video streaming and downloading, high volume business
applications and collaborative working, mobile internet and data transfer etc. are visible into the
high multimedia service type that supports a peak bit rate of up to 30 Mbit/s. Moreover, for the
average subscriber in developing countries, the medium multimedia service categories with peak
bit rate up to 2 Mbit/s are the most affordable multimedia services.
Table 5. ITU proposed radio parameters of RATG 1and RATG2
RATG1
Value
Macro Micro Pico
Cell
Cell
Cell
Application data rate
2010
20
40
40
(Mbit/s)
2015
20
40
40
2020
20
40
40
Max supported velocity-Km/h
250
50
4
Guard band between operators (MHz) 0
Minimum deployment per operators and RE (MHz) -Granularity of deployment per operator per RE (MHz) Number of overlapping network deploymentSupport for multicast
Attribute
Macro
Cell
0
50
50
250
RATG2
Value
Micro Pico
Cell
Cell
0
0
100
1000
100
1000
50
4
Hot
spot
0
1000
1000
4
120
120
40
20
1
yes
To ensure a high bit rate, operators need to install more base stations with radio frequency or
optical fiber based high capacity backhaul with supporting infrastructure. Since radio frequency is
a very scarce and limited resource, an optical fiber based transmission /backhaul network is
prepared to carry the required traffic of super high and high multimedia services. Apart from its
very high installation cost, operation and maintenance are also challenging in rural environments.
Since developing countries dont have a very big market, it remains difficult for operators to
establish economically viable and profitable network for IMT systems. Thus, for developing
countries we propose the application data rates for RATG 2 as given in Table 6 which can ensure
the optimum use of the network and reduce the capital expenditure needed.
Table 6. Proposed application data rate of RATG 2
Attribute
Application
data rate
[Mbit/s]
2010
2015
2020
Macro Cell
0
30
30
Value
Micro Cell
Pico Cell
0
0
50
100
50
100
Hotspot
0
100
100
21
The people of Bangladesh predominantly rely on traditional and relatively low- tech ICT
(Information Communication Technology) options to access information. In the ICT
development index released by ITU, Bangladesh ranked 138 out of 154 countries. Furthermore, a
large portion of the population suffers from digital divide and there is a big difference in socio22
economic conditions between the urban and rural populations. Due to the lack of basic literacy,
financial insolvency, computer skills and training in the use of ICT applications, it remains a
significant challenge to proliferate telecommunication services included in the IMT systems in
the rural areas of Bangladesh. Language barriers and the complexity of personal computer
operation are hindering the internet based service diffusion further. Moreover the relatively high
price of the 3G/4G/LTE terminal devices and the comparatively higher tariffs of high speed data,
multimedia and video telephony services may discourage lower and lower middle income people
from migrating from 2G to 3G/4G/LTE. It can be deemed that 40% of the total cellular mobile
subscriber will enter the 3G/4G/LTE network within the year of 2020. So there is a genuine need
to determine the spectrum requirements for Bangladesh that can accommodate not only the new
services of IMT systems but also the new radio transmission technologies being developed.
To estimate the spectrum requirement for IMT systems, the related market parameters are
provided in Report ITU-R M2072 for the years of 2010, 2015 and 2020. Those forecasts present
the extreme volume of the strongest and aggressive market of developed countries but cannot be
an example for spectrum estimation for IMT systems of a developing economy like Bangladesh.
ITU has applied market attribute settings on the forecasted user density, session arrival rate, mean
service bit rate and mean session durations [7]. For the years of 2010 and 2015 the user density is
5% in lower user density market settings and 25% for higher user density market setting while for
2020 the user density is 25% in lower user density market settings and 50% for higher user
density market setting for all service categories. Even after applying the lower user density
market setting on the market parameters of the Report ITU-R M2072, the ITU forecasted user
density is above the limit of the expected user density of Bangladesh for the years 2010, 2015 and
2020. For example we compare some typical market data of the most usable service categories
included in the conversational traffic class provided by ITU and the actual data that have been
collected from Bangladesh telecom market for the year 2010 in the Table 7. ITU forecasted user
density is 45 in the SE2 and 7 in the SE5 of the SC2 whereas the actual user densities are 20 and
3. Moreover there is quite a difference in the user density of the SC3 and SC5. This example is
also applicable for all other service categories and service environments for the years 2010, 2015
and 2020. For this reason the ITU forecasted spectrum estimation does not match with the
spectrum requirements of developing countries. Furthermore the session arrival rate per user and
average session durations of each service category in the sub-urban and rural environments are
larger than in the dense urban environments which indicate that the consumption rates in the suburban and rural environments exceed those of the dense urban environments for voice, data and
multimedia service. For this reason the spectrum demand is dominated by the sub-urban
teledensity in RATG1 [8]. This is another major limitation of the ITU spectrum estimation
methodology. The dense urban environment represents a central business and highly populated
area. Most users may use high-speed broadband data and multimedia services They also make
many phone calls from home, office and public areas. Hence in a dense urban environment, a
larger spectrum requirement holds than in a sub-urban environment. This scenario is also
applicable for Bangladesh because the user density of its dense urban areas is approximately
twenty times higher than the user density of sub-urban areas. So the market parameters have to be
changed to get reliable spectrum estimation results for IMT systems for Developing countries.
23
Table7. Typical market data (user density, session arrival rate mean service bit rate and average session
duration) for the year of 2010 (BGD indicates Bangladesh)
SC
SE
2
2
3
3
5
5
2
5
1
4
5
6
User Density
ITU
45
7
6107.7
1387.5
8095.5
677
BGD
20
3
3637.5
757.5
2228.5
380
Session Arrival
Rate
ITU
BGD
0.299
0.199
0.299
0.1
0.2437 0.2129
0.3738 0.158
1.688
0.511
1.3484 0.2506
BGD
10000
3000
496.2
290.2
16
16
Average Session
Duration
ITU
BGD
53
41
51
16
113.5
105.75
109.2
45.5
227.1
128.9
207.5
95.6
Note that the 2G/2.5G cellular mobile communication market of Bangladesh is in a saturation
state. For this reason we consider these service categories (SC4, SC5, SC9, SC10, SC14, SC15,
SC19 and SC20) with the higher user density market settings. To calculate the spectrum
requirement of Bangladesh we propose the user density parameter as shown in Table 8 based on
the cellular mobile, computer and internet penetration rates. Table 8 shows that user density
parameter denoted by U is equal to a certain percentage for different service categories. The
market setting for session arrival rate per user is denoted by Q, the average session duration by
and the mean service bit rate by R are equal to 30% for the year of 2010 and 2015, and 40% for
the year of 2020 in all service categories. The market setting for the mobility ratios presented in
this table assumes the middle mobility scenario for all service categories except for SC11.
Because SC11 is a very high data rate service category with more than 100 Mbit/s data rate
requirements.
Table 8. Market attributes settings for Bangladesh for the year of 2010, 2015 and 2020
SC
1
2
3
4
5
6
7
8
9
10
U in %
10
15
20
5
5
5
25
25
5
5
5
25
25
8
10
15
25
25
8
10
15
25
25
25
25
40
40
40
25
25
40
40
40
Q/R/
in %
10/ 20
15
30
30
30
30
30
30
30
30
30
30
40
40
40
40
40
40
40
40
40
40
Mob
-ility
ratio
SC
2
2
2
2
2
2
2
2
2
2
11
12
13
14
15
16
17
18
19
20
U in %
10
15
20
5
5
5
25
25
5
5
5
25
25
8
10
15
25
25
8
10
15
25
25
25
25
40
40
40
25
25
40
40
40
Q/R/
in %
10/ 20
15
30
30
30
30
30
30
30
30
30
30
40
40
40
40
40
40
40
40
40
40
Mobility
ratio
2
1
2
2
2
2
2
2
2
2
requirement at the cell edge. In that case two carriers could be deployed; one providing basically
the minimum ubiquitous service of 5 Mbps for all users and the other providing the access to
services demanding higher data rates as up to 50 Mbps per service. As an example two carrier
bandwidths of 2 x 20 MHz and 2 x 40 MHz are considered. 2x20 MHz operation could provide
an aggregate peak throughput between 40 60 Mbits, which would be sufficient to deploy basic
services in future wireless mobile systems. The 2 x 40 MHz carrier bandwidth could be deployed
in environments where users request services demanding higher data rates. The carrier bandwidth
of 2 x 40 MHz would allow a minimum of 80 Mbps that could guarantee mobility between cells
for the high data rate services [10]. The Asia Pacific Telecommunity (APT) spectrum information
database indicates that most of the countries in the Asia Pacific region have assigned 2x15 MHz
or 2x20 MHz spectrum to each operator for IMT systems, which can provide a minimum
ubiquitous data rate of 5 Mbps for all users in a full mobility radio environment.
5.
SPECTRUM
REQUIREMENT
ESTIMATION
FOR
BANGLADESH AND COMPARISON WITH THE FORECASTED
RESULT OF ITU
The complete algorithm developed by Hideaki Takagi and Bernhard H.Walke [8] and the
calculation tool package SPECULATOR [17] are used for traffic and spectrum requirement
estimation that has been described briefly in section 3 and 4. Calculations of spectrum
requirement estimates for Bangladesh are presented in Table-9 and demonstrated in fig.3 & 4.
These calculations were based on Bangladeshs telecom market trends, cellular mobile and
internet penetration rates. ITU presents the results of the calculation of spectrum requirements for
IMT systems in the Report ITU-R.M 2078 and the values are listed in Table-9 and illustrated in
fig. 5 & 6. In the present telecom scenario, most traffic has been shifting from voice oriented
communication to multimedia communication, so the spectrum requirement for RATG2 should
be higher than RATG1 for the year of 2020. But in table 10 & Fig.8, the ITU spectrum estimate
indicates that the spectrum requirement for RATG 2 is lower than RATG1. Moreover the
spectrum demand is dominated by the sub-urban teledensity of RATG1 instead of the dense urban
teledensity. These are the major shortcomings of the ITU methodology. In contrast, the results of
the spectrum estimate for Bangladesh show in table 8 & fig.7 that the spectrum requirement for
dense urban area is almost double than the spectrum requirement for sub-urban areas for every
forecasted year and that the spectrum requirement for RATG 2 is higher than RATG1 for the year
of 2020.
Table 9.Spectrum requirement estimation Result
RATG
RATG1
RATG2
Teledensity
Dense Urban
Sub-urban
Rural
Dense urban
Sub-urban
Rural
Spectrum requirement
estimation for Bangladesh
2010
2015
2020
520
700
540
300
360
340
160
220
200
0
340
680
0
200
400
0
140
180
25
RATG
RATG1
RATG2
520
0
700
340
540
680
2015
2020
840
0
880
420
880
840
Fig.7 Final spectrum requirement for Bangladesh Fig.8 Final Spectrum requirement forecasted by ITU
6. C ONCLUSIONS
In this paper, we analyzed the algorithm underlying ITUs methodology for calculating of
spectrum requirements of IMT systems. We proposed an alternate approach for estimating the
spectrum requirements for a developing country like Bangladesh to deploy IMT systems in the
three forecast years 2010, 2015 and 2020. We have pointed out the unsuitability of the ITU
26
methodology of applying spectrum estimation for developing countries and proposed the market
setting, cell area, population coverage percentage, application data rate, market parameters like
user density, mean service bit rate, session arrival rate and average session durations based on the
Bangladeshi mobile communications environment and market. According to the ITU-R spectrum
estimation methodology, the spectrum requirements of IMT systems for lower and higher user
density settings are 1160 MHz and 1720 MHz respectively for the year of 2020 [8].The predicted
total spectrum bandwidth requirements of Bangladesh for both the RATG 1 and RATG 2 for the
year 2020 are calculated as 1220 MHz. It should be noted that this figure (1220 MHz) is little
higher than the spectrum requirements of the lower market setting (1160 MHz). These
consequences indicate quite difference between the forecasting marks of ITU-R and the outcomes
based on the input parameters of Bangladesh telecom network deployment scenarios. Which
direct ITU spectrum estimation overestimates the approximate requirement of developing
countries.
REFERENCES
[1]
[2]
[3]
[4]
[5]
[6]
[7]
[8]
[9]
[10]
[11]
[12]
[13]
[14]
[15]
[16]
[17]
[18]
[19]
Rec. ITU-R M.1390, Methodology for the Calculation of IMT- 2000 Terrestrial Spectrum
Requirement, ITU, 2000.
Rep. ITU-R M.2023, Spectrum Requirements for IMT-2000, ITU, 2000.
Rec. ITU-R M.1645, Framework and Overall Objectives of the Future Development of IMT-2000 and
Systems beyond IMT-2000, ITU, 2003.
Rec. ITU-R M.1768, Methodology for Calculation of Spectrum Requirements for the Future
Development of IMT-2000 and Systems beyond IMT-2000 from the Year 2010 Onwards, ITU, 2006.
Rep. ITU-R M.2072, World Mobile Telecommunication Market Forecast, ITU, 2006.
Rep. ITU-R M.2074, Radio Aspects for the Terrestrial Component of IMT-2000 and Systems beyond
IMT 2000, ITU, 2006.
Rep. ITU-R M.2078, Spectrum Requirements for the Future Development of IMT-2000 and Systems
beyond IMT-2000, ITU, 2006.
Matinmikko, M. Huschke J. & Ojala, J. 2008. Calculation tool package. In: Tak-agi, H. & Walke, B.
H. (ed.) Spectrum requirement planning in wireless communications: Model and methodology for
IMT-Advanced. Chichester, UK: John Wiley & Sons.
Document ITU-R WP-5D/283-E, 2013, LM Ericsson, Intel Corporation, Nokia Corporation, Nokia
Siemens Networks: Proposed input parameter values for spectrum requirement estimation for IMT
systems.
WINNER System concept: Capabilities and Spectrum Usage, WWRF meeting 16, Shanghai, 26-28
April 2006.
Kleinrock L 1975 Queueing Systems, vol. 1: Theory. John Wiley & Sons
J.Kaufmann, Blocking in a shared resource environment, Transactions on Communications, vol.
COM-29, no. 10, pp. 14741481, October 1981.
H.Takagi, H.Yoshino, N.Matoba, and M. Azuma, Methodology for Calculation of Spectrum
Requirements for the Next Generation Mobile Communication Systems, IEICE Trans. on
Communications, vol. J89-B, no. 2, Feb. 2006, pp. 135-142.
A.Cobham,Priority Assignments in Waiting Line Problems Operations Research, vol. 2, pp. 7076,
1954.
H.Kesten and J. Runneberg, Priority in Waiting Line Problems, Proc. Koninkl. Nederlandse
Akademie van Wetenshappen, vol. 60, pp. 312324 and 325336, 1957, series A.
T.Irnich and B. Walke Spectrum Estimation Methodology for Next Generation Wireless Systems,
PIMRC 2004, 5-8 Sep. 2004, Barcelona, Spain.
SPECULATOR for estimating the spectrum requirements for IMT systems developed by ISTWINNER and WINNER II projects.
Chung, Lim, Yook and Park Calculation of spectral efficiency for estimating spectrum requirement
of IMT advanced in Korean Communication Environments ETRI journal, Volume 29, Number 2,
April 2007.
GSMA (2013) Spectrum required for various mobile communications markets in 2020.
27
Authors
Md Sohel Rana was born in Natore, Bangladesh in 1979. He received the B.Sc
degree in Electrical and Electronics Engineering in 2002 from Khulna
University of Engineering and Technology, Bangladesh. He has been working as
a Senior Assistant Director of Bangladesh Telecommunication Regulatory
Commission (BTRC) since 2004 and is responsible for spectrum planning and
management. Now he is in deputation at Kyung Hee University, Republic of
Korea for masters study. His research interest include radio spectrum
management, spectrum planning and interference analysis between wireless
communications systems.
Een-Kee Hong received the BS, MS and PhD degree in Electrical Engineering
from Yonsei University, Seoul, Republic of Korea in 1989, 1991 and 1995
respectively. Since 1999 he has been with the Department of Electronics and
Radio Engineering of Kyung Hee University. In addition of that he is working
as a director of Korea Information and Communication Society (KICS) and
Korea Navigation Institute (KONI). His research interest are wireless
communications (physical layer), spectrum engineering for cellular mobile
systems and cross layer optimization.
28
ISSN: 9081-2546
Journal of Distributed and Parallel Systems , Vol. 7, Issue 2, 2014
ABSTRACT
Communication in mobile Ad hoc networks is completed via multi-hop ways. Owing to the distributed
specification and restricted resource of nodes, MANET is a lot prone to wormhole attacks i.e. wormhole
attacks place severe threats to each Ad hoc routing protocol and a few security enhancements. Thus, so as
to discover wormholes, totally different techniques are in use. In all those techniques fixation of threshold
is merely by trial & error methodology or by random manner. Conjointly wormhole detection is in twin
part by putting the nodes that is higher than the edge in a suspicious set, however predicting the node as a
wormhole by using some other algorithms. Our aim in this paper is to deduce the traffic threshold level by
derivational approach for identifying wormholes in a very single phase in relay network having dissimilar
characteristics.
KEYWORDS
MANET, wormhole, Traffic prediction, parametric threshold implication, derivational approach.
1. INTRODUCTION
A mobile unintended network (MANET) could be a self configuration infrastructure less network
with non centralized administration. A mobile Ad hoc or unintended network is an autonomous
assortment of mobile devices (laptops, smart phones, sensors, etc.,)that communicate with one
another over wireless links and collaborate in a distributed manner so as to produce the required
network practically within the absence of a fixed infrastructure. Every device in a MANET is
unengaged to move independently in any direction, and can amend its links to different devices
often. This type of network, operating as a complete network or with one or multiple points of
attachments to cellular networks or the web, paves the approach for variable new and exciting
applications. Application scenarios embrace, however are not restricted to: emergency and rescue
operations, conferences are field settings, automobile networks, personal networking, etc. In
wormhole attack, malicious node receives knowledge packet at one point within the network and
tunnels them to a different malicious node. The tunnel existing between two malicious nodes is
named as a wormhole. The tunnel gets the information from one network and replicates to
different network. A wormhole therefore permits an attacker to form two attacker controlled
choke points which may be utilized by the attacker to degrade or analyze traffic at a desired point.
29
TPIDS is a lightweight traffic prediction intrusion detection theme that tumbling the value of
communication and energy by hastily detecting the behaviour of the intrusions. Some work has
been done to wormhole attack in mobile unintended networks however it cause excessive false
alarm rate. In this paper we are focusing on reducing warning rate by choosing optimum
threshold value to save lots of wastage of energy and information measure of mobile nodes in
sleuthing wormholes. The remaining components of this paper is organized as follows: section III
provides traffic prediction supported ARMA, section IV provides experimental analysis and
results, section V presents relation between attributes of the network and threshold values, section
VI presents threshold selection, finally conclusion is conferred in section VII.
2. RELATED WORK
The discussion starts by Yu Bos paper, that relies on the discovery of multi-hop recognition
theme to detect attacks caused by the selection of transmitting the irregular packet loss. In this
theme, a region of the transmission path, nodes are going to be randomly selected for testing.
Detection point is going to be generated for every incident packet to the upstream transmission
methods, any node within the middle, if not adequately recognized package, can generate warning
data of abnormal packet loss and to submit a multi-hop to the supply node. Here, we have
considered ton choosing transmission attack is taken into account, that introduces larger
communications and computing prices. Then khin sandar win [6] proposes solely an analysis of
detecting wormhole attack in wireless network, simply by quoting the benefits and drawbacks not
suggesting for the foremost economical one. Han zhijie[1] instructed traffic prediction
methodology, however he didnt decrease the false alarm rate whereas detecting wormholes.
Faizal M.A.[2] offers regarding solely a way to verify threshold values by using SPC approach
that are more necessary for detection of wormholes and conjointly to scale back false alarm rate
within the MANET. This observation is finished on real time network traffic having the aim of
distinguishing the typical connection created by the host or hosts to single victim among one
second interval.
Pij = P Xm+1 = j | Xm =i
(1)
Where
P ij The probability of entering the state j when a node is in the operational state i .
The migration probability of second-order is defined as P2ij that is, a node in the current state of
30
i will enter the state j after having two state transitions. (i.e.)
Where
can take any arbitrary values between 0 and n.
A further notation of Markov chain for probability is to use M*M matrix of P which is called as
migration probability matrix. In this matrix, the element Pij represents the probability in the i th
row and j th column.
Assuming that BS is the data transmissible data quantity of a node stays at state s. then after
calculating the number of domains, the total data transmissible quantity is given by the formula:
31
And also the data of each node in the time can be calculated by equation (7).the transmissible data
quantity for the total number of nodes in a cluster is given by
Where
Ck it is to represent ith node, in cluster Ck.
Pis(t) is the probability from i th state migrating to the sth state.
32
SCREEN SHOT 1:
This screen capture shows the combined execution of both prediction data algorithm and the
wormhole detection algorithm. Since the predicted data is got from the ARMA algorithm, it uses
probability matrix i.e the probability of a nodes to go from state 'i' to state 'j' after time t is
shown in the above result.
In order to deduce the relation between the threshold and network attributes first we have to
stumble on the relationship between the varying network characteristics and the normalized data.
For this we reflect on the data present in each node, number of nodes, time instants etc.
Varying the network attributes like amount of data in nodes, number of nodes, number of time
instants, the results have been simulated.
33
Predicted Data
1
2
3
4
5
6
2,2,2
2,4,6
5,10,15
20,30,40
60,70,80
50,100, 150
11.88001
28.20005
43.2000
129.60086
189.000031
180
This table (1) describes the relation between the predicted data and actual data in wormhole by
varying the amount of data in each node. The inference is that data in wormhole is more than data
in other nodes. More the increase in data in each node more will be the variation in predicted and
the actual one. With this variation into concern the below graph is plotted.
Fig. 1 Variance Vs Normalized data
The Fig 1 is plotted against variance calculated from the amount of data in each node, and the
normalized data calculated from the formula
Normalized data = Predicted Data Actual Data
Predicted Data
(9)
34
This graph infers that there is a diminishing relation between the amount of data in each node,
and the normalized data which is above calculated.
S.
No
No. of
nodes
Prediction
data
Actual
data in
worm
hole
Data in other
nodes
No.
of txm
20,30,40,50
1783.5
291
22, 51,110
20,30,40,50,60
5723.22
391
22, 51, 91
20,30,40,50,60,70
9007.2
551
22,52,91,110,101
20,30,40,50,60,70,80
25278
651
22,51,91, 111
20,30,40,50,60,
70,80,90
33282
811
22,51,91,111,131
20,30,40,50,60
5116.4
991
22,51,91,111
This table 2 infers that while varying the number of nodes a suitable relation is formed between
the predicted data and actual value of data from which a graph is plotted.
Fig. 2 No. of Nodes Vs Normalized data
35
This Fig. 2 infers that there is some linear relation between the varying nodes and the normalized
value calculated from the above equation (9) by tagging the varying number of nodes in the Xaxis.
36
5. RELATION BETWEEN
THRESHOLD VALUES
NETWORK
ATTRIBUTES
AND
Assume different threshold values and then experiment it with a network having characteristics
like number of nodes, traffic intensity, node density and also wormhole density. And then, find
false alarm rates for each assumed threshold values. The detection rate performance is highest for
one threshold value for this network of certain parameters. The experiment is then repeated by
varying the characteristics of the network and similar threshold values with best performance are
obtained. Such threshold values of highest detection rate performance are found for each network
with varying parameters. A mathematical relation is then deduced between the parameters of the
network and such obtained threshold values. Based on this relation, an optimal threshold value
with least false alarm rate can be selected for a network with any parameter set. This experiment
is simulated in NS2 and the expected result would be in the form of a mathematical expression
obtained by theoretical analysis.
6. THRESHOLD SELECTION
The normal and the abnormal traffic are differentiated using a threshold value. Thus suitable
selection and the correct threshold value add an extra advantage for IDS to detect anomalies in
the network. Selecting inaccurate threshold value will cause an excessive false alarm especially if
the value is too low or if it is too high, it can cause the intrusion activity being considered as
normal traffic. Most of the research does not propose a proper technique to identify the threshold
technique. Here threshold is determined by dynamic techniques. Dynamic threshold technique
requires prior knowledge of the network traffic before the threshold value can be selected.
7. CONCLUSION
In this paper, anomaly detection and security scheme based on Markov model is used by each
node in MANET to predict traffic (TPIDS). All the above analysis shows that there exists a
perceptible relation obtained by altering the network attributes, with this noticeable relation it is
evident that there prevails an optimal threshold based on this relation. On deriving this relation
wormhole detection can be ended in single phase. This relation is malleable to the protocol
37
DSDV which finds the wormhole by using shortest distance, hence in future work the optimal
threshold for other routing protocols such as AODV etc can be deduced.
REFERENCES
[1]
C Han Zhijie, Wang Ruchuang: Intrusion Detection for Wireless Sensor Network Based on Traffic
Prediction Model, 2012 International Conference on Solid State Devices and Materials Science.
[2] Faizal M. A., Mohd Zaki M., Shahrin S., Robiah Y, Siti Rahayu S., Nazrulazhar B.: Threshold
Verification Technique for Network Intrusion Detection System, (IJCSIS) International Journal of
Computer Science and Information Security,Vol. 2, No. 1, 2009.
[3] Ankita Gupta,Sanjay Prakash Ranga: Wormhole Detection Methods In Manet,International Journal
Of Enterprise Computing And Business Systems(IJECBS), Vol. 2 Issue 2 July 2012.
[4] Moutushi Singh, Rupayan Das: A Survey Of Different Techniques For Detection of Wormhole
Attack In Wireless Sensor Network , International Journal of Scientific & Engineering Research
Volume 3, Issue 10, October-2012 .
[5] Murad A. Rassam, M.A. Maarof and Anazida Zainal: A Survey of Intrusion Detection Schemes in
Wireless Sensor Networks American Journal of Applied Sciences, 2012.
[6] Khin Sandar Win, Pathein Gyi: Analysis of Detecting Wormhole Attack in Wireless Networks,
World Academy of Science, Engineering and Technology , 2008.
[7] Lukman Sharif and Munir Ahmed, The Wormhole Routing Attack in Wireless Sensor Networks
(WSN), Journal of Information Processing Systems, Vol.6, No.2, June 2010.
[8] Jing Deng, Richard Han, and Shivakant Mishra, Defending against Pathbased DoS Attacks in
Wireless Sensor Networks SASN05, November 7, 2005.
[9] Llker Demirkol, Fatih Alagoz,Hakan Delic, Cem Ersoy. Wireless Sensor Networks for
IntrusionDetection:PacketTrafficModeling[EB/OL].www.cmpe.boun.edu.tr/~ilker/IlkerDE
MIRKOL_COMML_ext_abstract . pdf.
[10] Onat I , Miri A. An intrusion detection system for wireless sensor networks/Proceedings of the IEEE
International Conference on Wireless and Mobile Computing , Networking and Communications
(WiMOB05) Mont real , Canada , 2005.
Authors
C.Susil Kumar received his B.E degree in Electronics and Instrumentation from
Madras University, India in 2001 and M.Tech degree in Communication Engineering
from Vellore Institute of Technology, India, in 2004. He is working as Assistant
Professor in the Department of Electronics and Communication Engineering,
Velammal College of Engineering and Technology, Madurai. His research interests
include Wireless Communication and Ad hoc Networks.
K.Aathi Dharshini is pursuing M.E Communication systems in Velammal College of
Engineering and Technology,Madurai.She received her B.E degree in R.V.S College
of Engineering and Technology, affiliated under Anna university, India, in 2012. Her
area of interest include Wireless Ad hoc Networks.
E.BabuThirumangaiAlwar received his B.E degree in Electronics and
Communication Engineering from Manonmaniam Sundaranar University, India in
1998 and M.E degree in Computer Science and Engineering from Anna University of
Technology, Tiruchirappalli, India, in 2010. He is working as Assistant Professor in
the Department of Computer Science and Engineering, Hindusthan Institute of
Technology, Coimbatore. His research interests include Wireless Communication
and Adhoc Networks.
38
ISSN: 9081-2546
Journal of Distributed and Parallel Systems , Vol. 7, Issue 2, 2014
ABSTRACT
The mobile WiMAX plays a vital role in accessing the delay sensitive audio, video streaming and mobile
IPTV. To minimize the handover delay, a Location Management Area (LMA) based Multicast and
Broadcast Service (MBS) zone is established. The handover delay is increased based on the size of the MBS
zone. In this paper, Location Management Area is easily identified by using Database Lookup Method to
obtain efficient bandwidth utilization along with reduced handover delay and increased throughput. The
handover delay and throughput is calculated by implementing this scenario in OPNET tool.
KEYWORDS
MBS zone, LMA, Paging Group, Database Lookup Method, Handover Delay.
1. INTRODUCTION
The most popular Internet Access technology for Metropolitan Area Networks (MAN) is
WiMAX (Worldwide Interoperability for Microwave Access), because WiMAX supports all
Internet Protocol core network architecture [6]. In WiMAX, recent research is going on in the
area of Broadband access which plays a vital role in mobile services. There should be a handover
delay in these sectors. Mostly, the handover delays are reduced in previous researches, although
its not more efficient at the users point of view, so the motivation is to reduce the bandwidth
usage and handover delay [15]. WiMAX is an emerging 4G technology, which can provide
wireless broadband access at the data rate of 70 Mbps over 50 kilometers to both home and
business customers. WiMAX is cost effective and offers higher data rates than other wireless
networks. It supports both fixed and mobile applications. WiMAX is easier to deploy as
compared to other networks and has flexible network architectures. One of the major challenges
concerning the performance of Mobile WiMAX is seamless handover. The handover in an MBS
session involves a delay, due to the link-level messages, which are exchanged during the mobile
WiMAX handover and due to the MBS signaling messages [4]. Whenever a Mobile Station with
an ongoing MBS session switches to a new Base Station, the former delay occurs, irrespective of
its MBS zone or LMA. The latter delay only occurs when an MS moves from one MBS zone to
another. MBS handovers can be classified into intra-MBS-zone handover and inter-MBS-zone
39
handover. Handover occurs within an MBS zone is mentioned as intra-MBS-zone handover and
inter-MBS-zone handover takes place between different MBS zones.
2. RELATED WORKS
The standard Internet Protocol (IP) multicast protocols are employed by Universal Mobile
Telecommunications System (UMTS) multicast architecture [9]. A multicast mechanism for
UMTS has been proposed [10] establishes a multicast tunnel throughout the UMTS network,
which allows multicast packets to be transferred on shared links towards the multiple
destinations. For one-to-many packet-delivery services in the UMTS, the tradeoffs between the
broadcast, multiple-unicast, and multicast approaches have been investigated [11]. However the
MBS zone creation will also waste the bandwidth of the wired-backhaul network, but we focus on
the wireless link so the bandwidth is also reduced. It is assumed that the standard UMTS mobility
mechanisms, handles the mobility. On the other hand [12], a routing list can be introduced into
the nodes of the UMTS to support resource-efficient multicast transmissions, which are combined
with a reassessment of the mobility management mechanism and handover types in the UMTS.
Unless handover-delay issues are taken into Account, however, multicast-service continuity still
cannot be assured. There is some ongoing research on the support of real-time services such as
video streaming (IP television) and voice over IP in mobile WiMAX [13], [14]. The hard
handovers by IEEE 802.16e make seamless mobility with imperceptible interruption of service
difficult to achieve in Mobile WiMAX. The aim of [2] is reducing both bandwidth usage and
service disruption. Some previous studies have addressed the issues of network planning for
wireless multicast and broadcast services. So they used paging method which reduces the service
disruption, but in this case the handover delay and the bandwidth are not much efficient. In this
paper, the Database Lookup method is introduced which is implemented in AAA server. By this
the bandwidth and handover delay is reduced and throughput is increased.
3. MBS ZONE
A WiMAX network basically consists of Mobile Stations, Base Stations, and gateways with an
authentication, authorization, and accounting server [5]. The group of Base Stations is named as
LMAs within an MBS zone. The LMA contains a group of base stations and mobile stations.
The MBS zones are controlled by the MBSC (Multicast and Broadcast Service Controller) which
has a direct link to the AAA server in that the Database Lookup LMA is implemented, as shown
in Figure 1. The MBS zones can be deployed, once the MBS-zone sizes are determined by a
network operator [3]. Every Base Station in an MBS zone has a shared multicast connection for
the same multicast transmission. Therefore, a Mobile Station to create a new connection during
handovers between Base Stations in the same MBS zone is not needed, which reduces the
handover delay and bandwidth usage. The service disruptions are reduced by increasing the size
of the MBS zones for a given level of mobility; hence the wireless-link bandwidth will be wasted
in the sense that every Base Station broadcasts the same MBS packets, in the same MBS zone
regardless of the presence of a user. This will also result in increased probability of blocking a
new MBS session.
With an LMA-based MBS, intra-MBS-zone handovers are again classified into intra-LMA
handovers, which result from a movement of the Base Station within the same LMA, and inter41
LMA handovers between Base Stations in different LMAs [3]. Figure 3 shows the signalling
messages required for inter-LMA handovers. LMA containing current mobile users of a specific
MBS session is called as active LMAs with a remainder called as inactive LMAs, with respect
to the session. In Figure 3, the Paging Controller (PC) updates location information about an
MBS user within the zone. The Paging Controller, which controls the paging group, informs the
MBSC of the new location of the users. If the MS has shifted into an inactive LMA, multicastdistribution tree update must be done. If the LMA is active, the time required for location update
and location update notification does not contribute to the handover delay, which only consists of
the time required to finish IEEE802.16e MAC layer handover. The delay involved in an intraLMA handover is equivalent to the intra-MBS-zone handover.
users of the particular area which serves under the respective AAA server. When the handover
occurs, it is not necessary to spend time in PG to collect the data of the subscribers. Directly, the
location updating can be done during MAC layer handover. So that the handover delay is reduced
and bandwidth utilization is decreased with increased throughput. In the proposed work, the MS
location is determined from a static database in the AAA server.
Once the network receives a location request, it identifies the cell ID of the BS serving the MS
and looks up the geolocation of that BS [5]. This provides the reduced handover delay with
efficient bandwidth usage.
The Figure 4 explains about the complete operation of this paper. In this figure, the operation is
between the MS, SBS (Serving Base Station), TBS (Target Base Station), MBSC, AAA (with
DLM). At First, the MS sends a IEEE802.16e MAC layer handover configuration to the TBS, the
TBS sends an MBS joint request to MBS Controller, which sends an MBS authorization request
to the AAA server. After the response from the AAA to the MBSC, if the user is valid then only
the MBSC sends a request to TBS and the TBS sends a DSA (Dynamic Service Addition) request
to MS. The MS sends a DSA response to the TBS only when the user is authorized and the DSA
acknowledgment is sent to the MS [8]. A Response is sent to MBSC from TBS. After the
resource reservation step, the location is updated to the AAA server from TBS. Then the location
update notification is sent to the TBS from the AAA server. Then normal process such as security
key exchange, MBS updating process, MBS data transmission is preceded.
43
The Database Lookup Method [5] for implementing LMA based MBS zone should enhance the
throughput of the network by the elimination of Paging Group and thus the handover delay is also
reduced efficiently. The bandwidth utilization is also effectively used in the removal of paging
controller.
44
5. SIMULATION RESULTS
OPNET simulator is a tool to simulate the behavior and performance of any type of network [7].
The main difference with other simulators lies in its power and versatility. This simulator makes
possible working with the OSI model, from layer 7 to the modification of the most essential
physical parameters [1]. The proposed work is implemented using this OPNET tool and the
results are obtained.
The Figure 5 is the scenario of existing work (LMA based MBS zone) in opnet simulator. It has
paging controller to track the MS location and the paging group is also created by grouping of
nearby Base Stations. The increased throughput and the reduced overall delay are calculated as in
Figure 6.
45
As in Figure 7, the handover delay for a particular node (node 6 and 22) is calculated while the
LMA handover takes place. The handover delay is calculated as 0.025 seconds, which is
46
maintained constantly in node 6. But, in node 22 the handover delay is increased from 0.020
seconds to 0.024 seconds.
The proposed scenario is created as in Figure 8 in the opnet tool which has the MBS zone and
group of LMAs. The Paging Group used to find the current position of the user is eliminated in
the proposed LMA based MBS zone [3]. Instead of that, a database lookup LMA is introduced, in
that the location of a MS is maintained in a database. The database is handled by the AAA server.
Hence the throughput is increased and the delay is minimized as compared to existing work. The
corresponding graph is shown in Figure 9.
48
The Figure 10 explains that the handover delay is reduced from 0.025 seconds to 0.021seconds
during node 6 mobility.Thus the handover delay is minimized due to the removal of Paging
Group in MBS zone. Similarly, during node 22 mobility also the handover delay is minimized
when compared to the value in Figure 7.
6. CONCLUSION
Mobile WiMAX (IEEE802.16e) includes the MBS zone to reduce the MBS service disruption
time and handover delays, but this requires all the BSs to send the same packets in the MBS zone.
The DLM implemented in AAA server has to maintain the updating of the mobile users. By using
LMA in MBS zone, the handover delay while streaming in the MBS zones and the service
disruption time will be reduced to an extent by maintaining the Database Lookup in the AAA
Server. The service-disruption time, which is dominated by the transmission delay between the
MBSC and the Base Station. If the LMAs are small, the disruption time of the LMS is hardly
affected to all by the wireless-transmission delay. In this paper the handover delay is calculated as
0.021 seconds, which is less than the handover delay in existing scenario.
REFERENCES
[1]
[2]
49
[3]
[4]
[5]
[6]
[7]
[8]
[9]
[10]
[11]
[12]
[13]
[14]
[15]
Lee, Ji Hoon, et al. "Location management area (LMA)-based MBS handover in mobile WiMAX
systems." Communication Systems Software and Middleware and Workshops, 2008. COMSWARE
2008. 3rd International Conference on. IEEE, 2008.
Becvar, Zdenek, and Jan Zelenka. "Handovers in the mobile WiMAX." Research in
Telecommunication technology 1 (2006): 147-50.
Venkatachalam, Muthaiah, et al. "Location services in WiMAX networks." Communications
Magazine, IEEE 47.10 (2009): 92-98.
WiMAX-Part, Mobile. "I: A technical overview and performance evaluation." WiMAX Forum. 2006.
Vukadinovic, Vladimir, and Ljiljana Trajkovic. "Mobile Application Part protocol implementation in
OPNET.
Lin, Zih-Ci, Huai-Lei Fu, and Phone Lin. "Dynamic Channel Allocation for Wireless Zone-Based
Multicast and Broadcast Service." INFOCOM, 2011 Proceedings IEEE. IEEE, 2011.
M. Hauge and O. Kure, Multicast in 3G networks: Employment of existing IP multicast protocols in
UMTS, in Proc. 5th ACM Int. Workshop.Wireless Mobile Multimedia, 2002, pp. 96103.
R. Rummler, Y.W. Chung, and A. H. Aghvami, Modeling and analysis of an efficient multicast
mechanism for UMTS, IEEE Trans. Veh. Technol., vol. 54, no. 1, pp. 350365, Jan. 2005.
A. Alexiou and C. Bouras, Multicast in UMTS: Evaluation and recommendations, Wireless
Commun. Mobile Comput., vol. 8, no. 4, pp. 463 481, May 2008. LEE et al.: REDUCING
HANDOVER DELAY IN MOBILE WiMAX MULTICAST AND BROADCAST SERVICES 617
A. Alexiou, D. Antonellis, and C. Bouras, An efficient mechanism for multicast data transmission in
UMTS, Wireless Pers. Commun., vol. 44, no. 4, pp. 455471, Mar. 2008.
S. Sengupta, M. Chatterjee, and S. Ganguly, Improving quality of VoIP streams overWiMAX,
IEEE Trans. Comput., vol. 57, no. 2, pp. 145156, Feb. 2008.
J. She, F. Hou, P.-H. Ho, and L.-L. Xie, IPTV overWiMAX: Key success factors, challenges, and
solutions, IEEE Commun. Mag., vol. 45, no. 8, pp. 8793, Aug. 2007.
W. Jiao, P. Jiang, and Y. Ma, Fast handover scheme for real-time applications in mobile WiMAX,
in Proc. IEEE Int. Conf. Commun., 2007.
Authors
S.Sivaranjani has received the Master of Engineering from Anna University, Chennai in
Communication Sysems. She is currently working as an Assistant professor in Theni
Kammavar Sangam College of Technology, Theni. She has attended International
Conferences and also published papers in international journal. She has research
interests in WiMAX security and Handover.
T.S.Supriya has received the Master of Engineering from Anna University, Chennai in
Communication Sysems. She is currently working as an Assistant professor in SCAD
Institute of Technology, Tirupur. Her research interests are in wireless networks and
security. She has also attended International Conferences and published paper in
international journal.
Dr.B.Sridevi, Professor of ECE Department of Velammal College of Engineering &
Technology, Madurai, obtained her B.E., degree from A.C.C.E.T Karaikudi , Madurai
Kamaraj University, Madurai and M.E. degree from Anna University, Chennai. She
has 2 years of Industrial experience, 10 years of Teaching,and Research experience.
Completed Ph.D. in Anna University, Chennai in the field of WiMAX Networks. She
have published many research papers in International journals, national and
international conferences. Her area of research includes Network Security, Wireless
Networks.
50
ISSN: 9081-2546
Journal of Distributed and Parallel Systems , Vol. 7, Issue 2, 2014
ABSTRACT
A Web page recommender system effectively predicts the best related web page to search. While searching
a word from search engine it may display some unnecessary links and unrelated datas to user so to avoid
this problem, the conceptual prediction model combines both the web usage and domain knowledge. The
proposed conceptual prediction model automatically generates a semantic network of the semantic Web
usage knowledge, which is the integration of domain knowledge and web usage information. Web usage
mining aims to discover interesting and frequent user access patterns from web browsing data. The
discovered knowledge can then be used for many practical web applications such as web
recommendations, adaptive web sites, and personalized web search and surfing.
KEYWORDS
Web Usage Mining, Ranking, Histories, Domain Knowledge, page recommendations.
1. INTRODUCTION
The main goal of this mining is used to find best link for users searching. Web usage mining is
the process of extracting knowledge from web users access by using data mining technologies.
This web usage mining application is called as recommender system. This recommender system
is to improve Web site usability.web usage mining prediction process is structured according to
web server activity and analyzing historical data such as server access log file or web logs which
are captured from the server then these web logs are used capturing the intuition list of the user so
as to recommend page views to the user whenever he/she comes online for the next time.
Our paper, we present architecture for capturing recommendations in the form of intuition list of
user. Intuition list consist of list of pages visited by user as well as the list of pages visited by
other user of having similar usage profile.
The results represent that improved accuracy of recommendations. The Web usage mining
process [6] consist of following three inter-dependent stages: collection of data, pre-processing,
pattern discovery and analysis. In the pre-processing stage, the click stream data is cleaned and
divided into a set of user transactions represents the behavior of each user during different
sessions. In the pattern discovery stage, statistical, database, and machine learning operations are
executed to get hidden patterns revealing the usual behavior of users, summary statistics on Web
resources, sessions, and users. In the final stage of the process, the extracted patterns and statistics
51
are further analyzed, filtered, which result in aggregate user models that is used as input to
applications such as recommendation engines, visualization tools, and Web analytics and report
generation tools. The overall process is depicted in Fig. 2.There is different types of models are
available.
1.1 Ontology
Ontology is describing the detailed information[1,5,7] from the domain data mining and
knowledge discovery it includes definition of basic data mining entities (e.g., data type, dataset,
data mining task, data mining algorithm etc.) and allows extensions with more complex data
mining entities (e.g. constraints, data mining scenarios and data mining experiments).
2. EXISTING SYSTEM
In an Existing System either ontology or semantic network model was used. The performance of
existing approaches depends on the sizes of training datasets. The bigger the training dataset size
is, the higher the prediction accuracy is. However, these approaches make Web-page
recommendations solely based on the Web access sequences [3] learnt from the Web usage data.
Therefore, the predicted pages are limited within the discovered Web access sequences.
Integrating semantic information with Web usage mining achieved higher performance than
classic Web usage mining algorithms. However, one of the big challenges that these approaches
are facing is the semantic domain knowledge acquisition and representation. Manually building
ontology of a website is challenging given the very large size [1] of Web data in todays websites.
So the performance of the system will be degraded.
3. PROPOSED SYSTEM
In this system using conceptual prediction model which combines the ontology model and
semantic network model Proposed system presents a new method to provide better Webpage
recommendation based on Web usage and domain knowledge, which is supported by three new
knowledge representation models and a set of Web-page recommendation strategies. The first
model is an ontology-based model [1] that represents the domain knowledge of a website. The
52
construction of this model is semi-automated so that the development efforts from developers can
be reduced. The second model is a semantic network [2] that represents domain knowledge,
whose construction can be fully automated. This model can be easily incorporated into a Webpage recommendation process because of this fully automated feature. The third model is a
conceptual prediction model, which is a navigation network of domain terms based on the
frequently viewed Web-pages and represents the integrated Web usage[2] and domain knowledge
for supporting Web-page prediction. The construction of this model can be fully automated.
The recommendation strategies make use of the domain knowledge and the prediction model
through two of the three models to predict the next pages with probabilities for a given Web user
based on the current Web-page navigation state.
4. SYSTEM ARCHITECTURE
Architecture describes about the process while searching a word in search engine. User gives the
query to the query processor, that query processor is to searching is based on 3 models. Ontology
model, Semantic network & Conceptual prediction model, Ontology contains user queries and
elaborated content. Semantic contains the relation between the data and corresponding result. By
combining these 2 models it has been proposed a conceptual prediction model based upon
filtering used to find the result set and also download ratio scheme is used to find the ranking
results based on content downloading. These 3 models based on following techniques
53
5. TECHNIQUES
In our Proposed Work Illustrates following techniques,
6. IMPLEMENTATION
Types to be describes are as follows,
6.1 Data Creation and Manipulations
6.2 User interface
6.3 Query Processing
6.4 Usage and Relationship mining
6.5 Ranking Model
54
7. SUMMARY
This paper illustrates, the related works on web usage mining process including web usage data,
preprocessing links, and the Sequence pattern construction techniques. Usage based data is the
main source for web usage mining; it mainly includes web server logs, proxy server logs and
client browser logs. they are the most widely used source in research on web usage mining. Web
search access patterns from websites. However, it also includes datas from user profiles,
registration details, cookies, user queries and bookmarks from the interactions of users while
surfing on the Web. Web usage data are mainly divided into three types, namely web server logs,
proxy server logs and client browser logs.
These paper techniques are generally used for extracting statistical knowledge from weblogs.
Such knowledge is most useful for analyzing web traffic of a website. Apriori technique can be
used for finding related pages that are most often referred together in an access session.
Clustering technique can be used to discover user clusters from web logs. Sequential patterns are
sequences of web pages accessed frequently by users. Such patterns are useful for discovering
user behavior and predicting future pages to be visited by the user.
8. CONCLUSIONS
A new web usage mining process for finding sequential patterns in web usage data which can be
used for predicting the possible next move in browsing sessions three new models has been
proposed. One is an ontology based model which defines domain knowledge. Second is semantic
network model which defines relationship and histories. A conceptual prediction model is also
proposed to integrate the Web usage and domain knowledge to form a weighted semantic
network. Results are filtered in this conception prediction model. That links are displayed in the
web page. These frequently used links only updated as a first link and also while downloading a
file and that link will be recommended in the web log as a first link and that is the best web page.
ACKNOWLEDGEMENTS
I dont have enough words to describe the profound gratitude and sense of indebtedness which I
feel to express towards my supervisor Mr.P.SABARINATHAN, Assistant Professor, Department
56
REFERENCES
[1]
Boyce S. and Pahl C.(2007) Developing Domain Ontologies for Course Content, Educational
Technology &Society,vol.10,pp.275-288.
[2] Dai M. and Mobasher B.(2005) Integrating Semantic Knowledge With Web Usage Mining for
Personalization,in Web Mining:Application And Techniques,Global,pp.276-306.
[3] Ezeife C.I. and Lu Y.(2005) Mining Web Log Sequential Patterns with Position Coded Pre-Order
Linked WAP Tree,Data Mining and Knowledge Discovery,vol.10,pp.5-38.
[4] Ezeife C.I. and Lu Y.(2009) Fast Incremental Mining of Web Sequential Patterns with PLWAP
Tree,Data Mining and Knowledge Discovery,vol.19,pp.376- 416.
[5] Eirinaki M ., Mavroeidi D ., Tsatsaronis G. and Vazirgiannis M.(2006) Introducing in Web
Personalization :The Role of Ontologies, Mining, pp.147-162.
[6] Liu B . , Mobashar B. and Nasraoui O.(2011) Web Usage Mining , in Web Data Mining: Exploring
Hyperlinks, Contents, and Usage Data,pp.527-603.
[7] Oberle D . ,Grimm S.and Staab S.(2009) An Ontology for Software ,in Handbook on Ontologies
vol.2 pp.383- 402.
[8] Rios S.A. and Velasquez J .D. (2008) Semantic Web Usage Mining by Concept - Based Approach
for Off-line Web Site Enhancements , in Web Intelligence and Intelligent Agent Technology,pp.
234-241
[9] Stumme G.,Hoth A.And Berendt B.(2004) Usage Mining for and on the Semantic Web,pp.461-480.
[10] Zhou B. (2004) Intelligent Web Usage Mining , Nanyang Technological University.
Authors
Brindha.S received her B.Tech degree in Information Technology from M.I.E.T
Engineering College, Tiruchirappalli in 2012. She is currently doing her ME-Computer
Science in Pavendar Bharathidasan College of Engineering and Technology,
Tiruchirappalli.
Sabarinathan.P received his BE degree in Computer Science from Annai Mathammal
Sheela Engineering College, Namakkal in 2007 and received his ME degree in the
same stream in 2010 from Dhanalakshmi Srinivasan Engineering College, Perambalur.
He is currently working as an Assistant Professor in Pavendar Bharathidasan College
of Engineering and Technology, Tiruchirappalli and his area of interest includes
MANET and Data mining.
57
ISSN: 9081-2546
Journal of Distributed and Parallel Systems , Vol. 7, Issue 2, 2014
ABSTRACT
In bound applications, the locations of events reportable by a device network have to be compelled to stay
anonymous. That is, unauthorized observers should be unable to notice the origin of such events by
analyzing the network traffic. I analyze 2 forms of downsides: Communication overhead and machine load
problem. During this paper, I gift a brand new framework for modeling, analyzing, and evaluating
obscurity in device networks. The novelty of the proposed framework is twofold: initial, it introduces the
notion of interval indistinguishability and provides a quantitative live to model obscurity in wireless
device networks; second, it maps supply obscurity to the applied mathematics downside I showed that the
present approaches for coming up with statistically anonymous systems introduce correlation in real
intervals whereas faux area unit unrelated. I show however mapping supply obscurity to consecutive
hypothesis testing with nuisance Parameters ends up in changing the matter of exposing non-public supply
data into checking out associate degree applicable knowledge transformation that removes or minimize the
impact of the nuisance data victimization sturdy cryptography algorithmic rule. By doing therefore, I
remodel the matter of analyzing real valued sample points to binary codes, that opens the door for
committal to writing theory to be incorporated into the study of anonymous networks. In existing work,
unable to notice unauthorized observer in network traffic. However our work in the main supported
enhances their supply obscurity against correlation check. the most goal of supply location privacy is to
cover the existence of real events.
KEYWORDS
source location, privacy, anonymity, consecutive hypothesis testing, sturdy cryptography algorithmic rule,
nuisance parameters.
1. INTRODUCTION
A Wireless Sensor networks is a network comprising of nodes that are connected wirelessly and
communicate with each other. Wireless sensor network remains one of the challenging research
domains. Sensor networks are used in several applications such as military, healthcare,
monitoring purposes and surveillance. Sensor networks provide time and location privacy and it
is better suited to physical environment. Nodes are used to transmit information when an event is
detected. SENSOR networks are deployed to sense, monitor, and report events of interest in a
very wide selection of applications as well as, however don't seem to be restricted to military,
health care, and animal chase. There are measure 3 parameters that may be related to an
occurrence detected and according by a device node: the description of the event, the time of the
event, and the location of the event.
59
Once device networks are deployed in untrusty environments, protective the privacy of the 3
parameters that may be attributed to an occurrence triggered transmission becomes a vital
security feature within the style of wireless device networks. The source anonymity downside in
wireless device networks is that the downside of learning techniques that offer time and
placement privacy for events according by device nodes. Time and placement privacy are used
interchangeably with supply anonymity throughout the paper.
Within the existing literature, the source namelessness problem has been self-addressed beneath 2
differing types of adversaries, namely, local and international adversaries. an area opponent is
outlined to be associate opponent having restricted quality and partial read of the network traffic.
Routing primarily based techniques are shown to be effective out of sight the locations of
according events against native adversaries. A global soul has the ability to monitor the traffic of
the complete network (e.g., coordinative adversaries spatially distributed over the network).
Against international adversaries, routing primarily based techniques are to be ineffective in
concealing location data in event-triggered transmission. this is often as a result of the actual fact
that, since a worldwide soul has full abstraction read of the network, it will straight off sight the
origin and time of the event triggered transmission. We have a tendency to introduce the notion of
interval identicalness and illustrate how the matter of applied math supply namelessness will be
mapped to the matter of interval indistinguishability. Nodes are deployed to sense events of
interest and report them with minimum delay. Consequently, given the placement of a definite
node, of the placement, the node of the rumored event of interest will be approximated. Within
the nodes communication vary at the time of transmission. When a node senses an occasion, it
places data regarding the event in a very message and broadcast encrypted version of the
message. To obscure the report of an occasion of interest, nodes square measure assumed to
broadcast faux messages, even though no event of interest has been detected.
2. LITERATURE SURVEY
In this paper [2], Preserving source location privacy is becoming one of the most interesting
problems in wireless sensor networks. In a variety of real life applications, such as the
deployment of sensor nodes in battle fields, the locations of events monitored by the network are
required to remain anonymous. Given the knowledge of the network topology, however, an
adversary can expose the locations of such events by determining the individual nodes reporting
them. Here real events must be delayed until the next scheduled Fake transmission. When real
events have time-sensitive information, such latency might be unacceptable.
In paper [3], SPINS has two secure building blocks: SNEP and TESLA. SNEP provides the
following important baseline security primitives: Data confidentiality, two-party data
authentication, and data freshness. Particularly hard problem is to provide efficient broadcast
authentication, which is an important mechanism for sensor networks. TESLA is a new protocol
which provides authenticated broadcast for severely resource-constrained environments. SPINS
does not place any trust assumptions on the communication infrastructure, except that messages
are delivered to the destination with nonzero probability.
In paper [4], a routing technique is used to provide adequate source-location privacy with low
energy consumption. We introduce this technique as the Sink Toroidal Region (Star) routing.
With this technique, the source node randomly selects an intermediate node within a designed
60
Star area located around the SINK node. The Star area is large enough to make it unpractical for
an adversary to monitor the entire region. Furthermore, this routing protocol ensures that the
intermediate node is neither too closes, nor too far from the SINK node in relations to the entire
network. Here Network traffic and passive attack is possible.
In paper [5], they propose two techniques that prevent the leakage of location information:
periodic collection and source Simulation. Periodic collection provides a high level of location
privacy, while source simulation provides trade-offs between privacy, communication cost, and
latency. Through analysis and simulation, we demonstrate that the proposed techniques are
efficient and effective in protecting location information from the attacker. Communication
problem and network traffic is the major drawback here.
In paper [6],they first formulate this privacy problem as a constrained optimization problem and
then develop heuristics for an efficient privacy algorithm. Using simulations with randomized
movement models we verify that the algorithm improves privacy while minimizing the
perturbation of location samples.
61
Here the inputs are taken in a sequential manner to improve the security. Then determine the
correlation value and map the value to sequential hypothesis testing to remove the redundant
information. For providing source location privacy, I use robust encryption algorithm to send the
message to the target recipient without any interruption. In this approach, I am going to improve
security and perform network security analysis in an efficient manner to reduce latency time.
Here fake transmission is impossible.
In figure 4(a),it consists of single input and single output. Here the information is transferred
from transmitter to receiver. Each base station receives the signal from transmitter and sent to the
destination. Here we consider existing system as well as proposed system. In existing system,
Anderson test and binary hypothesis testing is used. But in our approach, we detect the source
anonymity problem with the help of sequential hypothesis testing as well as protect our system
from hackers using robust encryption algorithm.
62
Figure shows the block diagram for detection. Here the information is continuous that the real
values are converted into the binary values for identifying the noise parameters. Next we have to
find the correlation value for the input data. With the help of the correlation value, it undergoes
hypothesis testing to determine which the data sample belongs to. With the help of testing
hypothesis, we have to determine the transmission is real or fake. If it is a fake node, attacker will
be there. Here we are using sequential hypothesis testing. With the help of the above diagram, we
are going to detect the hacker transmission. After detection, we have to protect our encrypted
message from hackers.
next we describe the correlation value to set the threshold value that will be used for experimental
analysis of source anonymity based on sequential hypothesis testing.
It can be verified that the value of correlation should be in the range 0 to 1. The goal of the SPRT
is to decide which hypothesis is correct as soon as possible (i.e., for the smallest value of k). To
do this the SPRT requires two thresholds, 1 > 0. The SPRT stops as soon as k> 1, and we then
decide H1 is correct, or when k<=0, and we then decide H0 is correct. The key is to set the
thresholds so that we are guaranteed a certain levels of error. Making 1 larger and 0 smaller yields
a test that will tend to stop later and produce more accurate decisions. We will try to set the
thresholds to provide desired probabilities of detection PD and false-alarm PFA.
With the help of SPRT, we are going to detect attacker in the system. If we perform hypothesis
testing, we need test statistic value,. If the test statistic value is greater than 0.01, then the network
will be more secure otherwise there may be a possibility of attacker. Using hypothesis testing, we
are going to detect the hacker in our wireless sensor network. Here we can overcome the problem
of computational overhead and communication overhead. By using sequential hypothesis testing,
we have to reduce source anonymity problem.
Figure 5(a) shows the mechanism of protecting our information from hackers. For that we require
public key encryption and identity based encryption. Here the message is encrypted and identity
based encryption is used to identify the target recipient. Here we consider the target recipient as
sensor node. Then the message will reach to destination. Here in prevention, packet broadcasting
is most important factor in wireless sensor networks. Here messages are divided into packet and
packets are encrypted and send in a sequential manner. The information will send to the particular
target recipient using identity based encryption.
6. RESULT ANALYSIS
In wireless sensor networks, first we create a node between source and destination. Nodes are to
be active when the information is transmitted from source to destination. Nodes are
communicated in a wireless medium. We have to create any number of nodes.
64
Mesh topology is used for secure communication compare to all other network topology. Here we
are having detection and protection techniques. The experimental results showed that how to
create nodes and how the message is transferred from source to destination and the methodology
used for detection.
65
In fig 6(b), the message is encrypted and the message is divided into packets and the packets are
sent in a sequential manner in order to reduce the communication overhead and computational
load problem. The message can reach the destination in secure manner.
In the above graph, message can sent from source to destination. In between them, sensor nodes
between them. Message can be transmitted in a wireless medium. In sensor networks, there may
be a possibility of attack. This attack can results in modification of message, masquerade and
disclosure. In order to detect hacker in this system, we go for sequential hypothesis testing in
order to design an efficient system. in our next step, we have to provide source location privacy
and to reduce the network traffic. Secure communication is possible.
Fig 6(c): detection of hacker node when information is transmitting from source to destination.
For finding the shortest path we use djikistra algorithm and detect there may be any hacker node
using sequential hypothesis testing. In the above graph, we have to identify the number of fake
nodes during transmission. This solution is merely presented how to reduce source anonymity
problem based on sequential hypothesis testing. Using the proposed framework, including the
mapping of source anonymity to sequential hypothesis testing in order to design the notion of
interval indistinguishability. Fake events are identified. The main goal is to reduce latency and to
66
reduce network traffic. SPRT is the only choice to detect fake event. Hence the attacker is unable
to hack the original information .our system must be more secure by considering correlation tests.
7. CONCLUSIONS
In this paper, we provided a framework for source anonymity based on sequential based
hypothesis testing. We introduce the notion of interval indistinguishability to provide source
location privacy. By mapping the problem of source anonymity to sequential hypothesis testing in
order to design an efficient system. We showed why previous studies are unable to detect
attacker. Finally we propose a modification in the existing system to reduce network traffic.
REFERENCES
[1]
[2]
[3]
[4]
[5]
[6]
[7]
Basel Alomair, Member, Andrew Clark, Jorge Cuellar, and Radha Poovendran, Towards Statistical
framework for source anonymity in wireless sensor networks, IEEE TRANSACTIONS ON
MOBILE COMPUTING, VOL. 12, NO. 2, FEBRUARY 2013
B. Alomair, A. Clark, J. Cuellar, and R. Poovendran, On Source Anonymity in Wireless Sensor
Networks, Proc. IEEE/IFIP 40th Intl Conf. Dependable Systems and Networks (DSN 10), 2010.
A. Perrig, R. Szewczyk, J. Tygar, V. Wen, and D. Culler, SPINS: Security Protocols for Sensor
Networks, Wireless Networks, vol. 8,no. 5, pp. 521-534, 2002.
K. Mehta, D. Liu, and M. Wright, Location Privacy in Sensor Networks against a Global
Eavesdropper, Proc. IEEE 15th Intl Conf. Network Protocols (ICNP 07), pp. 314-323, 2007.
B. Hoh and M. Gruteser, Protecting Location Privacy Through Path Confusion, Proc.
IEEE/CreatNet First Intl Conf. Security and Privacy for Emerging Areas in Comm. Networks
(SecureComm 05), pp. 194-205, 2005.
N. Li, N. Zhang, S. Das, and B. Thuraisingham, Privacy Preservation in Wireless Sensor Networks:
A State-of-the-Art Survey, Elsevier J. Ad Hoc Networks, vol. 7, no. 8, pp. 1501-1514,2009
M. Shao, Y. Yang, S. Zhu, and G. Cao, Towards Statistically Strong Source Anonymity for Sensor
Networks, Proc. IEEE INFOCOM, pp. 466-474, 2008.
Author
Uma.G was studying M.E-Communication Systems in Parisutham
Institute Of Technology and Science,Thanjavur,Tamilnadu, India.
(Affliated to Anna University).
67
ISSN: 9081-2546
Journal of Distributed and Parallel Systems , Vol. 7, Issue 2, 2014
ABSTRACT
Scheduling different types of packets such as real-time and non-real time data packets in wireless links is
necessary to reduce energy consumption of the wireless device. Most of the existing packet scheduling
mechanism uses opportunistic transmission scheduling, in which communication is postponed upto an
acceptable time deadline until the best expected channel conditions to transmit are found. This algorithm
incurs a large processing overhead and more energy consumption. In this paper we propose a Dynamic
Multilevel Queue Scheduling algorithm. In the proposed scheme, the ready queue is partitioned into three
levels of priority queues. Real-time packets are placed into the highest priority queue and non-real time
data packets are placed into two other queues. We evaluate the performance of the proposed Dynamic
Multilevel Queue Scheduling scheme through simulations for real-time and non-real time data. Simulation
results illustrate that the Multilevel Priority packet scheduling scheme overcomes the conventional methods
interms of average data waiting time and end-to-end delay.
KEYWORDS
Dynamic Multilevel Priority (DMP), end-to-end delay, average task waiting time.
1. INTRODUCTION
Wireless Networks have evolved a lot and there is a necessity for energy consumption of the
wireless devices and we need to manage the network resources. Wireless devices cannot have an
uninterrupted power supply. In recent wireless communication networks, the major issue is the
decrease of the transmission power consumption. The channel circumstances are time-variant in
wireless atmosphere. The Wireless channel experiences Small-scale fading due to Multipath and
large-scale fading due to Shadowing. Scheduling of packets at the data transmission part is very
much essential as it ensures deliverance of diverse data packets based on their priority. Real-time
data packets have very high priority when compared to non-real time data packets. At present,
most of the Wireless Networks operate using First Come First Served (FCFS) scheduling
algorithm that transfers the packets according to their arrival time and it needs more time to be
transmitted to a Base Station (BS). Anyway, the data packet must reach the Base Station before
the deadline or within a particular time period. And also within minimum end-to-end delay, the
real-time data must be delivered to BS.
In this paper, we propose Dynamic Multilevel Queue Priority Scheduling algorithm where the
nodes are organized into a hierarchical structure. Nodes which have equal hop distance from the
BS are said to be present at the same hierarchical level. Each node upholds three levels of Priority
Queues, since we organize data packets as a) real-time (Priority-1) b) non-real time data packets
69
from lower level nodes (Priority-2) c) non-real time data packets present at the node itself
(Priority-3). Shortest Job First (SJF) scheduler is being used to process the non-real time data
packets that are present at the same level priority.
The remaining of this paper is organized as follows. Section II discusses various Related Works.
Section III presents a review of numerous existing Packet Scheduling algorithms. Section IV
describes various postulations and terminologies of DMP packet scheduling scheme. Section V
presents the experimental results and finally Section VI concludes the paper.
2. RELATED WORKS
The energy consumption problem in Wireless Networks has attracted world-wide. Many works
[1-3] has been done with a vision of minimizing the energy consumption on the wake-up mode in
wireless systems. A real-time architecture for large-scale networks [26] was proposed where
priority based scheduler is used. The data packets which travel maximum distance from the
source node to BS and have the minimum deadline are prioritized. A packet scheduling scheme
and algorithm called RACE [27] for real-time large scale networks was proposed. It uses
Bellman-Ford algorithm inorder to find out ways with less traffic and delay. Earliest Deadline
First (EDF) scheduling algorithm was used in RACE to transmit packets with shortest deadline.
[29] presents the mostly used operating system of Wireless Network and differentiate them as Cooperative and Preemptive. Co-operative scheduling algorithms are based on Adaptive Double
Ring Scheduling (ARDS) and EDF [30], that has two queues with various priorities. Based upon
the deadline of the arriving packets , the scheduler switches between the two queues. Cooperative schedulers are used in applications with limited resources. Preemptive Scheduling is
based on EFRM scheme which is an extension of Rate Monotonic (RM) scheme. In [6] , the state
of distributed data aggregation in Wireless Networks is being reviewed.
70
at the queue are of equal size. Data aggregation is not carried out at the intermediate nodes for
real-time packets. Nodes are assumed to be situated at different levels on the basis of number of
hops from Base Station.
Time Division Multiple Access scheme is being employed to allot timeslots to nodes at various
levels. Nodes at the lowest level are allotted timeslot1. There are three levels of priority in the
ready queue such as real-time data (priority 1) and Non-real time data (priority 2 and 3).
The length of the priority level queues is variable. Priority 1 level queue is shortest. The length of
both priority 2 and priority 3 level queues is same.
4.2 TERMINOLOGIES
4.2.1.Routing Protocol: We use Zone-based routing protocol for better energy efficiency and for
stability in energy consumption among the nodes. In a zone-based routing protocol, each zone
contains a Zone Head (ZH) for identification and their structure is based upon the number of hops
they are far-off from the Base Station (BS). Nodes in the zones that are one hop distant from the
BS is considered to be at level 1 and similarly those which are at two hop distant from the BS is
considered to be at level 2. Each zone is subdivided into smaller number of squares such that a
node in square envelops all other neighboring squares.
4.2.2.TDMA Systems: Task scheduling at each level is performed using TDMA method. Data
packets are sent from the lower level nodes to the BS via intermediate nodes. Comparing to the
lower-level nodes , nodes present at the top and intermediate levels have more processing
conditions. The time-slot of lower-level nodes is set to lowest length contrast to the higher value
of length of the timeslot of upper-level nodes. Intermediate level nodes must be stopped from
aggregating data as they have to be transmitted to the users with possibly lesser delay.
4.2.3. Fairness: This metric guarantees that tasks of diverse priorities are done with least waiting
time on the basis of priority of chore at the ready queue. For example, if any lower-priority chore
waits for more time for the constant income of higher-priority chores, fairness defines a limit that
permits the lower-priority tasks to be processed following a definite waiting time.
i. Real-time emergency packets with highest priority to accomplish the overall aim of the
Wireless Networks
ii. Non-real time packets to accomplish minimum average waiting time and end-to-end delay
iii. Non-real time packets with lowest priority to accomplish fairness.
5. PERFORMANCE ESTIMATION
The simulation model has been executed by using MATLAB. In DMP packet scheduling scheme,
one of the priority levels for non-real time data packets use Dynamic RR scheduling algorithm.
This is used to reduce average waiting time and end-to-end delay. The performance is evaluated
for the DMP packet scheduling algorithm, contrast to Dynamic Round Robin scheduling concept.
This comparison is done in terms of end-to-end transmission delay and average packet waiting
time. The number of simulated zones differs from 4 to 12 zones. Nodes are allocated identically
over the zones. The ready queue at each node can hold utmost 50 tasks. Type ID is used inorder
to identify its type.
73
Fig 1 & 2 demonstrate the end-to-end data transmission delay of real-time tasks over a number of
zones and levels, respectively. In both cases, we examine that DMP using Dynamic RR
outperforms the traditional DMP packet scheduler. This is due to the highest priority given to the
74
real-time tasks and since it allows the real-time packets to preempt the process of non-real time
data packets. Real-time packets have lesser transmission delay.
Fig 3 illustrates that the DRR packet scheduler has better performance compared with the DMP
packet scheduler interms of average task waiting time.
6. CONCLUSION
In this paper, we discuss about Dynamic Multilevel priority packet scheduling method. This type
of scheduler uses three priority levels. Here, the ready queue gets separated into three levels of
priority queues. Higher priority has been given to the processing of real-time data packets. In our
proposed method, we modify one of the scheduling scheme for processing non-real time data
packets and we apply Dynamic RR scheme in the second or third priority level of the Dynamic
Multilevel queue. This is done in order to minimize the average waiting time of the data packets
and also minimizes end-to-end delay. Experimental results show that the DMP using DRR
outperforms the normal DMP scheme interms of end-to-end delay and average data waiting time.
REFERENCES
[1]
[2]
[3]
[4]
[5]
[6]
[7]
[8]
[9]
[10]
[11]
[12]
[13]
[14]
R. Zheng, J. C. Hou, and L. Sha, Asynchronous wakeup for ad hoc networks, in Proc. 4th ACM Int.
Symp. Mobihoc, Annapolis, MD,Jun. 2003, pp. 3545.
X. Yang and N. H. Vaidya, A wakeup scheme for sensor networks: Achieving balance between
energy saving and end-to-end delay, in Proc.IEEE 10th RTAS, Toronto, Canada, May 2004, pp. 19
26.
M. Peng and W. Wang, An adaptive energy saving mechanism in the wireless packet access
network, in Proc. IEEE WCNC, Las Vegas, NV, Apr. 2008, pp. 15361540.
R. Yu, Y. Zhang, Z. Sun, and S. Mei, Energy- and QoS-aware packet forwarding in wireless sensor
networks, in Proc. IEEE ICC, Glasgow, U.K., Jun. 2007, pp. 32773282.
A. O. Fapojuwo, Energy consumption and message delay analysis of QoS-enhanced base-stationcontrolled dynamic clustering protocol for wireless sensor networks, IEEE Trans. Wireless
Commun., vol. 8, no. 10, pp. 53665374, Oct. 2009.246810121400.20.40.60.811.21.41.61.82x 104No
of zonesReal time Task End to End delay DMPDRR
Z. Ye, A. A. Abouzeid, and J. Ai, Optimal stochastic policies for distributed data aggregation in
wireless sensor networks, IEEE/ACM Trans. Netw., vol. 17, no. 5, pp. 14941507, Oct. 2009.
G. Anastasi, M. Conti, and M. Di Francesco, Extending the lifetime of wireless sensor networks
through adaptive sleep, IEEE Trans. Industrial Informatics, vol. 5, no. 3, pp. 351365, 2009.
G. Bergmann, M. Molnar, L. Gonczy, and B. Cousin, Optimal period length for the CQS sensor
network scheduling algorithm, in Proc. 2010 International Conf. Netw. Services, pp. 192199.
P. Guo, T. Jiang, Q. Zhang, and K. Zhang, Sleep scheduling for critical event monitoring in wireless
sensor networks, IEEE Trans. Parallel Distrib. Syst., vol. 23, no. 2, pp. 345352, Feb. 2012.
F. Liu, C. Tsui, and Y. J. Zhang, Joint routing and sleep scheduling for lifetime maximization of
wireless sensor networks, IEEE Trans.Wireless Commun., vol. 9, no. 7, pp. 22582267, July 2010.
J. Liu, N. Gu, and S. He, An energy-aware coverage based node scheduling scheme for wireless
sensor networks, in Proc. 2008 International Conf. Young Comput. Scientists, pp. 462468.
O. Khader, A. Willig, and A. Wolisz, Distributed wakeup scheduling scheme for supporting periodic
traffic in wsns, in Proc. 2009 European Wireless Conf., pp. 287292.
B. Nazir and H. Hasbullah, Dynamic sleep scheduling for minimizing delay in wireless sensor
network, in Proc. 2011 Saudi International Electron., Communications Photon. Conf., pp. 15.
D. Shuman and M. Liu, Optimal sleep scheduling for a wireless sensor network node, in Proc. 2006
Asilomar Conf. Signals, Syst. Comput., pp. 13371341.
75
[15] S. Paul, S. Nandi, and I. Singh, A dynamic balanced-energy sleep scheduling scheme in
heterogeneous wireless sensor network, in Proc. 2008 IEEE International Conf. Netw., pp. 16,
2008.
[16] A. R. Swain, R. C. Hansdah, and V. K. Chouhan, An energy aware routing protocol with sleep
scheduling for wireless sensor networks, in Proc. 2010 IEEE International Conf. Adv. Inf. Netw.
Appl., pp. 933940.
[17] N. Edalat, W. Xiao, C. Tham, E. Keikha, and L. Ong, A price-based adaptive task allocation for
wireless sensor network, in Proc. 2009 IEEE International Conf. Mobile Adhoc Sensor Syst., pp.
888893.
[18] H. Momeni, M. Sharifi, and S. Sedighian, A new approach to task allocation in wireless sensor actor
networks, in Proc. 2009 International Conf. Computational Intelligence, Commun. Syst. Netw., pp.
7378.
[19] W. Stallings, Operating Systems, 2nd edition. Prentice Hall, 1995.
[20] F. Tirkawi and S. Fischer, Adaptive tasks balancing in wireless sensor networks, in Proc. 2008
International Conf. Inf. Commun. Technol.: From Theory Appl., pp. 16.
[21] X. Yu, X. Xiaosong, and W. Wenyong, Priority-based low-power task scheduling for wireless sensor
network, in Proc. 2009 International Symp. Autonomous Decentralized Syst., pp. 15.
[22] Y. Zhao, Q. Wang, W. Wang, D. Jiang, and Y. Liu, Research on the priority-based soft real-time
task scheduling in TinyOS, in Proc. 2009 International Conf. Inf. Technol. Comput. Sci., vol. 1, pp.
562565.
[23] TinyOS. Available: http://webs.cs.berkeley.edu/tos, accessed June 2010.
[24] Available: http://webs.cs.berkeley.edu/tos, accessed June 2010.
[25] E. M. Lee, A. Kashif, D. H. Lee, I. T. Kim, and M. S. Park, Location based multi-queue scheduler in
wireless sensor network, in Proc. 2010 International Conf. Advanced Commun. Technol., vol. 1, pp.
551555.
[26] C. Lu, B. M. Blum, T. F. Abdelzaher, J. A. Stankovic, and T. He, RAP: a real-time communication
architecture for large-scale wireless sensor networks, in Proc. 2002 IEEE Real-Time Embedded
Technol. Appl. Symp., pp. 556
[27] K. Mizanian, R. Hajisheykhi, M. Baharloo, and A. H. Jahangir, RACE: a real-time scheduling policy
and communication architecture for largescale wireless sensor networks, in Proc. 2009 Commun.
Netw. Services Research Conf., pp. 458460.
[28] M. Yu, S. J. Xiahou, and X. Y. Li, A survey of studying on task scheduling mechanism for TinyOS,
in Proc. 2008 International Conf. Wireless Commun., Netw. Mobile Comput., pp. 14.
[29] P. A. Levis, TinyOS: an open operating system for wireless sensor networks (invited seminar), in
Proc. 2006 International Conf. Mobile Data Manag., p. 63.
[30] K. Lin, H. Zhao, Z. Y. Yin, and Y. G. Bi, An adaptive double ring scheduling strategy based on
tinyos, J. Northeastern University Natural Sci., vol. 28, no. 7, pp. 985988, 2007.
Authors
R.Arasa Kumar was born on 15.11.1985. He completed his B.E. in Electronics and
Communication Engineering in 2007 and his M.E. in Communication Systems in
2009. He has teaching experience of about 5yrs. At present, he is working as an
Assistant Professor in Velammal College of Engineering and Technology, Madurai,
Tamilnadu, India. He has published 3 papers in International Journals. His area of
interest includes Wireless Communication.
K.Madhu Varshini completed her B.E. in Electronics and Communication in 2012 and
now doing her final year M.E in Velammal College of Engineering and Technology,
Madurai. Her areas of interest include Wireless Networks and Optical
Communication. She has published a paper in International Conference.
76
ISSN: 9081-2546
Journal of Distributed and Parallel Systems , Vol. 7, Issue 2, 2014
ABSTRACT
Todays most modern research area of computing is cloud computing due to its ability to diminish the costs
associated with virtualization, high availability, dynamic resource pools and increases the efficiency of
computing. But still it contains some drawbacks such as privacy, security, etc. This paper is thoroughly
focused on the security of data of multi tenant model obtains from the virtualization feature of cloud
computing. We use AES-128 bit algorithm and cloud SQL to protect sensitive data before storing in the
cloud. When the authorized customer arises for usage of data, then data firstly decrypted after that
provides to the customer. Multi tenant infrastructure is supported by Google, which prefers pushing of
contents in short iteration cycle. As the customer is distributed and their demands can arise anywhere,
anytime so data cant store at particular site it must be available different sites also. For this faster
accessing by different users from different places Google is the best one. To get high reliability and
availability data is stored in encrypted before storing in database and updated every time after usage. It is
very easy to use without requiring any software. This authenticate user can recover their encrypted and
decrypted data, afford efficient and data storage security in the cloud.
KEYWORDS
Cloud computing, Multi tenant, AES, Cloud SQL, Google App Engine.
1. INTRODUCTION
In this modern period internet is working as a conventional hosting system which is accessible
through different services with limited usage and storage. But the current drift in business
requires vitality in computing and storage, it causes the development of cloud computing. For the
issues of computing, storage, and software, cloud computing proposes new models. This model
provides expansion in the environment, allocation and reallocation of assets when desired, storage
and networking ability virtually. It satisfies on demand need of the customers. Cloud computing
work as a combination of computational paradigm and distribution architecture. The major aim is
to provide services like net computing services, flexible data storage containing required
resources visualized as a service and delivered over the internet [1] [2]. Cloud computing
enhances many features like scalability, availability, adapt progress of demand, teamwork, pick
up the pace of development, provide potential for outlay fall through efficient and optimized
computing [3][ 4].
One of the important characteristics of cloud computing is multitenancy. This feature is similar in
nature with other multiple families on the same platform. It contained large no. Of resources and
data of different users which get differentiate by their unique identification. In this multitenancy
model certain level of control is provides for customizing and tailoring of hardware and software
for fulfilling customer demands. In multi tenancy model physical server partition occur with
virtualization. This virtualization feature contains good capability of separation. But still some
security issues arise those are data isolation, architecture expansion, configuration self definition
77
and performance customization. This feature favours users to copy, create, drift and roll back
virtual machines for running many applications [5] [6]. Like physical machine security, virtual
machine security is also important Virtual machine because error in any virtual machine may
cause in error in other [7]. Virtual machine as two boundaries- physical and virtual as physical
servers [8]. This paper is basically focused on data isolation.
From the aspect of quality of service data security is important (cong et al., 2009). When
the customer arises to satisfy their demand, it firstly assures that their data will be secure. Due to
which we use encryption algorithm to secure data storage and access. For this AES (advanced
encryption standard) 128 bit encryption algorithm is used to encrypt data before storing data in a
database. When the customer came to use resources or data on the physical server if it is
authorized then firstly decrypted and after that it provides to the user. Multi tenant infrastructure
is supported by many Applications, Google is one of them. Google pushes the contents in short
iteration. There are large of functionalities are added weekly basis, due to which required update
also
done
weekly
basis
in
a
Google
cloud
(http://www.Google.com/Apps/intl/en/business/cloud.html). When the new features are
introduced, they automatically reflected in the browser. As Google sustain by cloud computing, it
gets updated to fulfil millions of customer demands. The customers needs are not only
maintained on single site, but other secured centres also, so that when one site fails to fulfil the
customer demand it get fulfil by another site. Google supports parallel and fast access from
various places. For flexibility and reliability data are stored in various data centres. It is easy to
use Google cloud SQL. It does not require any type of software. Google cloud SQL concerns, my
SQL instances, which is similar to my SQL. It contains all features as my SQL and other
contained features are:
Instances limited up to 10GB, synchronous duplication, importation database, export database,
command line tools, highly obtainable, completely managed, SQL prompt.
In this study, we suggest a way of implementing the AES algorithm on the data before storing in
the database by the cloud service provider (CSP). Every cloud service seeker should confirm the
security criteria with cloud security provider before hosting data in the database. As many tenants
cause many changes to their data after every regular interval, so there are many challenges to
overcome.
providing multitenant architecture which basically deals with the security issues related to the
tenants data from each other. In SAAS or organizational data is serving in plain text and then
stored in the cloud. So SAAS provider is one dependable for security of data at the time of
serving and storing [13]. There are basically two issues of data security are as follows:
A. Security of data at transmit state.
B. Security of data at rest state.
C. Security of data in transmit state is basically concerned with isolation at network level.
While security of data at rest state is further divided in four parts where data is residing and
security issues can arise.
A. Application level security.
B. User level security, concerned with mobility and replication of data of users.
C. Virtual level security, concerned with unauthorized exposure, unapproved migration, physical
compromise.
D. Tenant level security, concerned with the data between the tenants.
3. P ROPOSED W ORK
Multi tenancy is the model where the physical server provides with partition by virtualization.
This partitioned server refers as a virtual machine (VM) and the users become tenants. Basically
data of tenants are stored in the database, so design of the database is also plays an important role.
Typically three ways are available to store data:
This paper basically focuses on the third concept of separate rows. In the multi tenant model
tenants data isolation is a great issue. So from the security point view separate database is
provided to each of the tenants, but it is less reliable and also time taking. To overcome this
penalty a single database is used by all the tenants on the same physical machine. But still there
are problems of security of data of tenants from each other which get reduced by providing the
concept of cryptography. Through cryptography concept we use AES (advanced encryption
standard) algorithm to encrypt data of the tenant before storing in the database. Each row of
database separates by their differentiating by their ids.
Each time when the customer comes to store their data, it firstly faces the encryption boundary of
AES algorithm provided by the cloud service provider (CSP). Where it gets encrypted and after
that it store in database.
4. B ASE METHODOLOGY
AES algorithm- basically AES is a symmetric block algorithm. This means encryption and
decryption is done by the same key. This algorithm can accept block size of 128 bits and use three
keys of choice, i.e. 128, 192, 256 bits Based upon which type of version is used by the customer;
hence the standard is named as AES-128, AES-192, and AES-256 respectively. The encryption
process of AES consists of 10 rounds of processing for 128 bits. All rounds are identical except
the last one. AES algorithm begins with a stage of adding round key followed by 9 rounds of four
stages and tenth round of three stages. The required stages are as follows:
79
1. Sub bytes- In this step of sub bytes a look up table is used in which each byte is replaced with.
2. Shift rows- In this step of shift rows, shifting of rows occurs cyclically with a particular offset,
while leaving the first unchanged.
3. Mix column- In this step of mixing operations occur using an invertible linear conversion in
order to merge the four bytes in each column.
4. Add round key- This step includes derivation of round keys from Rijindaels key agenda and
this round key gets added to each byte of the state.
In these last steps are performed up to fourth round, fifth round contains above all step except mix
column step as shown in figure 1.
4. EXPERIMENTAL METHODOLOGY
There are following steps which require implementing the AES algorithm in cloud to create
Google Application:
Step 1: firstly enter the URL http://accounts.Google.com/, then Google user name and password.
Step 2: Select Google Application link (my Application).
Step 3: Choose Create Application key, grant Application identifier, Application heading and
then get on Create Application button. At this instant Application is ready.
Now we implement AES algorithm in the Google cloud:
There are subsequent methods to create database, tables in Google cloud SQL
Step 1: Enter https://code.Google.com/apis/console and go for Google Cloud SQL alternative.
Step 2: Select New Instance tab from the right upper corner and popup window displayed.
Step 3: Now enter the instance name and correlated an authorative Application which is formed
earlier and then click on Create Instance button.
80
Step 6: For creating a database for the Application we use Create Database query, all necessary
tables are created.
Step 7: Add record to the desired table, we use Insert Into query.
Step 8: Generate client interface for the Application.
Step 9: Now write Java code of the AES algorithm to implement algorithm in cloud and debug
the Application in Google cloud.
Step 10: Encrypted get sore and decrypted data is displayed while accessing.
81
5. SAMPLE DATA
82
A User Interface and Application are Created Using Java and Jsp in Eclipse by Accquiring
Following Steps are:
Step 1: Database is created in Google cloud named as Studententry.
Step 2: Student Entry Detail table is created in student entry database with its required field like
name, id, contact department etc.
Step 3: An application Student entry was created in Google app engine by applying above
specified steps shown in figure2.
Step 4: To operate the Student Entry Details we designed user interface. User interface uses these
details. It firstly choose student link, then Student Entry Detail is displayed as shown in Figure 3
and Figure 4.
Step 5: Now by clicking on Submit button the details is received by student class and keys are
generated using AES algorithm.
Step 6: Using these generated keys data is encrypted using AES algorithm and then stored in
database.
Step 7: At the stage of data retrieval it is decrypted using generated keys.
83
6. C ONCLUSIONS
In our proposed work we mainly focus on data security of tenants so that only the authorized user
can access the data. To provide this feature we use encryption and decryption process of data so
that the only legal tenant can access their particular data. Hence, for security related to data of
tenants we implement AES (advanced encryption standard) using cloud SQL. The result obtained
from experimental methodology proved that AES gives protection for the data stored in the cloud.
We make use of AES algorithm and Google App Engine to supply secured data storage,
efficiency, assure availability in the condition of cloud denial-of-service attacks and data security
in the cloud. This Approach is basically implemented by tenants who are going to store their data
in the cloud. This approach is implemented by tenant itself at the cloud security provider (CSP)
who stores data in the database.
ACKNOWLEDGEMENTS
This research paper is made possible through the help and support from everyone, including:
parents, teachers, family, friends, and in essence, all sentient beings.
REFERENCES
[1]
[2]
[3]
[4]
[5]
S. Zhang, S. Zhang, X. Chen, and X. Huo, "Cloud Computing Research and Development Trend," In:
Second International Conference on Future Networks (ICFN 10), Sanya, Hainan, China IEEE
Computer Society ,Washington, DC, USA, pp. 93- 97, 2010.
G. Zhao, J. Liu, Y. Tang, W. Sun, F. Zhang, X. Ye, and N. Tang, "Cloud Computing: A Statistics
Aspect of Users," In: First International Conference on Cloud Computing (CloudCom), Beijing,
China. Springer Berlin, Heidelberg, pp. 347 - 358, 2009.
Marinos. A and G. Briscoe, "Community Cloud Computing," In: 1st International Conference on
Cloud Computing (CloudCom, Beijing, China. Springer-Verlag Berlin, Heidelberg, 2009.
A. Khalid, "Cloud Computing: applying issues in Small Business," In: International Conference on
Signal Acquisition and Processing (ICSAP 10), pp. 278 - 281, 2010.
A. Jasti, P. Shah, R. Nagaraj, and R. Pendse, "Security in multi-tenancy cloud," In: IEEE International
Carnahan Conference on Security Technology (ICCST ) , KS, USA. IEEE Computer Society,
Washington, DC, USA, pp. 35- 41, 2010.
84
[6]
[7]
[8]
[9]
[10]
[11]
[12]
[13]
T. Garfinkel and M. Rosenblum, "When virtual is harder than real: Security challenges in virtual
machine based computing environments," In: Proceedings of the 10th conference on Hot Topics in
Operating Systems, Santa Fe, NM. volume 10. USENIX Association Berkeley, CA, USA, pp. 227229, 2005.
L. Ertaul, S. Singhal, and S. Gkay, "Security challenges in Cloud Computing," In: Proceedings of the
2010 International conference on Security and Management SAM 10. CSREA Press, Las Vegas, US,
pp. 36- 42, 2010.
M. Morsy, J. Grundy, and I. ller, "An analysis of the Cloud Computing Security problem," In:
Proceedings of APSEC 2010 Cloud Workshop. APSEC, Sydney, Australia, 2010.
B. Grobauer, Walloschek. T, and E. Stocker, "Understanding Cloud Computing vulnerabilities," IEEE
Security Privacy 9(2), pp. 50-57, 2011.
J. Rittinghouse and J. Ransome, "Security in the Cloud," In: Cloud Computing. Implementation,
Management, and Security, CRC Press, 2009.
S. Subashini and V. Kavitha, "A survey on Security issues in service delivery models of Cloud
Computing," Journal of Network Computer Application pp. 1-11, 2011.
J. Viega, "Cloud Computing and the common Man," IEEE Computer Society , 42(8), pp. 106 - 108,
2009.
J. Ju, Y. Wang, J. Fu, J. Wu, and Z. Lin, "Research on Key Technology in SaaS," In: International
Conference on Intelligent Computing and Cognitive Informatics (ICICCI), Hangzhou, China, IEEE
Computer Society, Washington, DC, USA, pp. 384- 387, 2010.
Authors
Parul Kashyap is student of M.Tech., Dept. of Computer Science and Engg.,
S.R.M.U., Lucknow, Uttar Pradesh, India and she has completed B.Tech.(Computer
Science & Engineering) from RIET Kanpur, India in 2012. Currently, She is working
research on Data Security. She has contributed in various research papers in various
National and International journals.
Rahul Singh, Asst. Professor, Dept. of Computer Science and Engineering, SRMU,
Lucknow, Uttar Pradesh, India. He has completed M.Tech (Computer Science &
Engineering) from Motilal Nehru National Institute of Technology, Allahabad, India
and B.Tech (Computer Science & Engineering) from Ajay Kumar Garg Engineering
College, Ghaziabad, India 2013 and 2010 respectively. His research interests include
Cloud computing, Distributed Computing and Mobile computing.
85
ISSN: 9081-2546
Journal of Distributed and Parallel Systems , Vol. 7, Issue 2, 2014
ABSTRACT
MANET is an autonomous system of mobile nodes attached by wireless links. It represents a complex and
dynamic distributed systems that consist of mobile wireless nodes that can freely self organize into an adhoc network topology. The devices in the network may have limited transmission range therefore multiple
hops may be needed by one node to transfer data to another node in network. This leads to the need for an
effective routing protocol. In this paper we study various classifications of routing protocols and their types
for wireless mobile ad-hoc networks like DSDV, GSR, AODV, DSR, ZRP, FSR, CGSR, LAR, and Geocast
Protocols. In this paper we also compare different routing protocols on based on a given set of parameters
Scalability, Latency, Bandwidth, Control-overhead, Mobility impact.
KEYWORDS
MANET, Routing Protocol, Reactive Proactive, Hybrid, Hierarchical, Geographic Position, Assisted
Routing Protocol.
1. INTRODUCTION
Mobile Ad-hoc Network (MANET) is collection of wireless nodes that do not depend on already
existing infrastructure so there is no concept of base station or access point. In MANETs, due to
availability of mobility in nodes and deficiency of centralized entity, the network topology
changes repeatedly and erratically [1]. In MANETs each node works as router for packet
forwarding whereas in wired network router performs routing table. It is multi-hop wireless
network because different sets of nodes want to establish a network & it is not compulsory that
each node is within the transmission range as it might be in out of range, so another set of nodes
are used to connect the out of range nodes. Therefore whenever one node sends data to another
node, a set of nodes may be used in between, where data is send in different hop thats why they
are also called multi-hop, wireless & distributed network [2].
Figure 1 depicts a multi-hop data transfer in a MANET. In the ad-hoc network nodes act as
routers as well as hosts therefore node may forward packets as well as run user applications [3,
4].The aim of MANET is to establish an accurate and efficient route between nodes such that any
messages are delivered on time [5]. Nowadays, with the immense growth in wireless network
applications like PDAs and cell phones, various researches are being done to improve the
network services and performance. So there are various challenging design issues in wireless
Ad Hoc networks [6].
Main challenges of these networks are:
Spectrum allocation
Self configuration
Medium access control
Energy efficient
Mobility management
Security & Privacy
Routing protocols
QoS etc
Main applications of this network are home network, environmental monitoring and public
wireless access in urban area, Emergency rescue and Vehicular communications in military.
packet reaches the destination. It typically uses distance-vector routing algorithms that keep
information about next hops to adjacent neighbours [17, 18].
2.2.2. Dynamic Source Routing (DSR)
Dynamic Source Routing is a Reactive Protocol based on the concept of source routing in which
source initiates route discovery on demand basis in multi-hop networks. The sender determines
the route from source to destination and it also includes the address of all intermediate nodes from
source to destination to the route record in the packet. Also called as a beaconless protocol where
HELLO messages are not exchanged between nodes to inform them about the presence of their
neighbours in the network [19].There are two key phases in DSR: route discovery and route
maintenance. All nodes maintain route caches that contain the source routes of which the mobile
is aware. The route caches entries are continually updated as new routes are learned. Route cache
is checked first when a source node wants to send a packet. If the route is available, the source
node incorporates the routing information inside the data packet before sending it [20].
90
fisheye technique to reduce the size of information required to represent graphical data. It helps
to make a routing protocol more scalable by assembly data on the topology [31, 32]. Distance is
calculated by hops from the node and is used to classify zones in FSR.
2.4.2. Cluster Gateway Switch Routing Protocol (CGSR)
CGSR is a multi-hop mobile wireless network with various routing schemes [15] in which nodes
are organized into hierarchy of clusters. It is a multichannel operation capable protocol [33]
where each node has a cluster head and packet is sent through cluster heads. Cluster heads
communicate amongst themselves using DSDV and two clusters are connected through a gateway
node. A packet sent by a source node is first sent to its cluster head, and then the gateway receives
packet from the cluster head. The gateway then sent it over to another cluster head, and this
process goes on until the cluster head of the destination node is reached. A cluster head is able to
control a group of ad-hoc hosts and each node maintains two tables, first table is: cluster member
table that contain the cluster head for each destination node & second table is: DV-routing table
that contain the next hop to the destination.
Figure 6. CGSR
92
3. P ERFORMANCE C OMPARISONS
Since there are number of routing protocols and their different algorithms as discussed above
therefore there is a need to compare different routing protocols to judge the performance and their
usage over different networks. The comparison done here is based on a given set of parameters
such Scalability, Latency, Bandwidth, Control-overhead, Mobility impact.
Table 1. Comparison of Various Routing Protocol
93
94
4. C ONCLUSIONS
We have seen a great improvement in the field of wireless and Mobile ad hoc network .In this
paper we have described a number of algorithms for routing and broadly categorized routing
protocols- Table driven, on demand, Hierarchical, Hybrid and Geographic position assisted
routing protocols and compared the various routing protocol of mobile ad-hoc networks and
presented in the form of table for a given parameter. There is not any defined single protocol that
can be perfect for usage in all type of networks. For comparatively small network proactive and
reactive routing protocols are appropriate. But in large network can be either hierarchical or
geographic routing protocols are suitable.
REFERENCES
[1]
[2]
[3]
[4]
[5]
[6]
[7]
[8]
[9]
[10]
[11]
[12]
[13]
[14]
[15]
[16]
[17]
Elizabeth Royer, and Chai-Keong Toh, (1999), A Review of Current Routing Protocols for Ad Hoc
Mobile Wireless Networks, IEEE Personal Communications, pp 46-55.
Robinpreet Kaur and Mritunjay Kumar Rai, (2012), A Novel Review on Routing Protocols in
MANETs, Undergraduate Academic Research Journal (UARJ), vol. 1, Issue-1, ISSN : 2278
1129.
P. Johansson, T. Larsson, N. Hedman, B. Mielczarek, M. Degermark, (1999), Scenario-based
Performance Analysis of Routing Protocols for Mobile Ad-hoc Networks, ACM Proceedings of the
5th annual International Conference on Mobile Computing and Networking, pp.195-206, isbn 58113-142-9.
Prabhakara Reddy, E. Suresh Babu, and M. L. Ravi Chandra, (2012) A comprehensive study of
Routing protocols in Mobile Ad hoc Networks, Research Survey International Journal of Advanced
Information Science and Technology (IJAIST), ISSN: 2319:2682, Vol.7, No.7.
Elizabeth M. Royer and Chai-Keong Toh, (1999) A Review of Current Routing Protocols for Ad
Hoc Mobile Wireless Networks, IEEE Personal Communications, vol. 6, no. 2.
Dr.S.S.Dhenakaran and A.Parvathavarthini, (2013) An Overview of Routing Protocols in Mobile
Ad-Hoc Network, International Journal of Advanced Research in Computer Science and Software
Engineering, Volume 3, Issue 2.
Larry L. Peterson and Bruce S. Davie, (1996) Computer Networks - A Systems Approach, Morgan
Kaufmann Publishers Inc., San Francisco, ISBN l-55860-368-9.
Robinpreet Kaur, and Mritunjay Kumar Rai, (2012), A Novel Review on Routing Protocols in
MANETs, Undergraduate Academic Research Journal (UARJ), Vol. 1, Issue-1, ISSN: 2278 1129.
Elizabeth M. Royer, (1999) A Review of Current Routing Protocols for Ad Hoc Mobile Wireless
Networks , IEEE Personal Communications, pp. 46-55.
P. Manickam, T. Guru Baskar, M. Girija and Dr. D. Manimegalai, (2011), "Performance comparisons
of routing protocols in mobile ad hoc networks", International Journal of Wireless & Mobile
Networks (IJWMN), vol. 3, no. 1.
Adel.S.El ashheb, (2012), "Performance Evaluation of AODV and DSDV Routing Protocol in
wireless sensor network Environment ", International Conference on Computer Networks and
Communication Systems, PCSIT vol.35, pp. 55-62.
Perkins CE, and Bhagwat P, (1994), Highly Dynamic Destination-Sequenced Distance-Vector
Routing (DSDV) for Mobile Computers, Proceedings of ACM SIGCOMM, pp. 234244.
T. Chen, and M. Gerla, (1998), Global State Routing: A new Routing Scheme for Ad-Hoc Wireless
Networks, Proceedings of IEEE ICC, pp. 171 - 175.
Krishna Gorantala, (2006), Routing Protocols in Mobile Ad-hoc Networks, A Masters thesis in
computer science, pp. 1-36.
Tarek Sheltami and Hussein Mouftah, (2003), Comparative study of on demand and Cluster Based
Routing protocols in MANETs, IEEE conference, pp. 291-295, 2003.
Charles E. Perkins, Elizabeth M. Belding-Royer, and Ian D. Chakeres, (2004) Mobile Ad Hoc
Networking Working Group INTERNET DRAFT.
Ian D.Chakeres, and Elizabeth M.Belding-Royer, (2004) AODV Routing Protocol Implementation
Design International Conf. on Distributed Computing Sysmtes, IEEE, vol.7.
95
[18] G.Vijaya Kumar, Y.Vasudeva Reddyr, and Dr.M.Nagendra, (2010), Current Research Work on
Routing Protocols for MANET: A Literature Survey, International Journal on Computer Science
and Engineering, Vol. 02, No. 03, pp. 706-713.
[19] S. Meenakshi Sundaram,Dr.S.Palani, and Dr. A. Ramesh Babu, (2013) A Performance Comparison
study of Unicast and Multicast Routing Protocols for Mobile Ad hoc Networks, International Journal
of Engineering Research and Applications (IJERA) , Vol. 3, Issue 2, pp.1446-1458.
[20] S. Sathish, K. Thangavel and S. Boopathi, (2011), "Comparative Analysis of DSR, FSR and ZRP
Routing Protocols in MANET ",International Conference on Information and Network Technology ,
IPCSIT vol.4, IACSIT Press.
[21] Ravi Prakash, Andre Schiper and Mansoor Mohsin, (2003) Reliable Multicast in Mobile Networks,
Proc. of IEEE, (WCNC).
[22] Dr.S.S.Dhenakaran , A.Parvathavarthini, (2013), "An Overview of Routing Protocols in Mobile AdHoc Network",International Journal of Advanced Research in Computer Science and Software
Engineering,Volume 3, Issue 2.
[23] M.L Sharma, Noor Fatima Rizvi, Nipun Sharma, Anu Malhan and Swati Sharma, (2010)
Performance Evaluation of MANET Routing Protocols under CBR and FTP traffic classes, Int. J.
Comp. Tech. Appl., vol. 2 (3),pp. 392-400.
[24] Pearlman MR, and Samar P, (2002 ) The Zone Routing Protocol (ZRP) for Ad Hoc Networks,
IETF draft, available at http://tools.ietf.org/id/draft-ietf-manetzone-zrp-04.txt. Accessed.
[25] Chiang C-C, Wu H-K, and Liu W, Gerla M, (1989) Routing in Clustered Multihop, Mobile
Wireless Networks with Fading Channel, Proceedings of IEEE SICON,pp.197211.
[26] Zygmunt J. Haas, and Marc R. Pearlman, (2001) The performance of Query Control Schemes for the
Zone Routing Protocol, IEEE/ACM transactions on networking, vol. 9, no. 4.
[27] A.-S.K. Pathan, C.S. Hong Haas ZJ, Pearlman MR, Samar P, (2002)Intrazone Routing Protocol
(IARP), IETF Internet Draft.
[28] Weiliang Li, and Jianjun Hao, (2010) Research on the Improvement of Multicast Ad Hoc Ondemand Distance Vector in MANETS, IEEE International Conference on Computer and Automation
Engineering (Singapore), vol. 1, pp.702-705.
[29] Mehran Abolhasan and Tadeusz Wysocki, (2003) Displacement based Route update strategies for
proactive routing protocols in mobile ad hoc networks International Workshop on the Internet,
Telecommunications and Signal processing.
[30] Pei G., Gerla M., and Tsu-Wei Chen, (2000)"Fisheye State Routing: A Routing Scheme for Ad Hoc
Wireless Networks," IEEE ICC vol. 1, pp. 70 -74.
[31] A. Iwata, C.C. Chiang, G. Pei, M. Gerla, and T.W. Chen, (1999) Scalable Routing Strategies for
Ad- Hoc Wireless Networks, IEEE Journal on Selected Areas in Communications, pp. 17(8): 1369 1379.
[32] Mario Gerla, Xiaoyan Hong, and Guangyu Pei, (2002) Fisheye State Routing Protocol (FSR) for Ad
Hoc Networks, INTERNETDRAFT-<draft-ietf-manet-fsr-03.txt>.
[33] C.C. Chiang, H.K. Wu, W. Liu, M. Gerla, (1997) Routing in Clustered Multihop Mobile Wireless
Networks with Fading Channel, Proceeding of IEEE Singapore International Conference on
Networks SICON, pages 197 -212.
[34] Biao Zhou,Yeng-Zhong Lee,Mario Gerla, (2008) Direction Assisted Geographic Routingfor Mobile
Ad Hoc networks ", IEEE Military communications conference (San Diego), pp. 1-7.
[35] Divya Wadhwa,Deepika,Vimmi Kochher and Rajesh Kumar Tyag, (2014) "A Review of Comparision
of Geographic Routing Protocols in Mobile Adhoc Network ",ISSN 2231-1297, Volume 4, Number 1
, pp. 51-58.
[36] Konglin Zhu , Biao Zhou, Xiaoming Fu, and Mario Gerla, (2011) Geo-assisted Multicast InterDomain Routing (GMIDR) Protocol for MANETs, IEEE International Conference.
[37] S.Esakki Muthu, and S.Sudha, (2013) Dynamic Geographical broadcast over Mobile ad Hoc
networks, IJREAT International Journal of Research in Engineering & Advanced Technology, Vol.
1, Issue 1, ISSN: 2320 8791.
96