Anda di halaman 1dari 5

PRIORITY BASED MULTIPLE BUFFER MANAGEMENT

FOR WEIGHTED PACKETS


1. A.Rabiyathul basariya M.E (Second Year), 2. Mrs. P.Parvathi M.Tech.,
Sudharsan Engineering College

Abstract arriving packets. Therefore, some packets


already in the buffer might have to be
Motivated by providing differentiated dropped. Without loss of generality, time
services on the Internet, we consider efficient is assumed to be discrete. In each time
online algorithms for buffer management in step, the buffer management selects a
network switches. The online buffer pending packet in the buffer to send. The
management problem formulates the problem buffer management can be regarded as an
of queuing policies of network switches online scheduling algorithm processing
supporting QOS (Quality of Service) prioritized packets. Fig. 1 illustrates the
guarantee. In FIFO buffering model, in which buffer structure inside a network switch.
unit-length packets arrive in an online
manner and each packet is associated with a
value (weight) representing its priority. The
order of the packets being sent should comply
with the order of their arriving time. The
buffer size is finite. At most one packet can be
sent in each time step. The paper proposes the
concept of multi-buffer management for
efficient packet delivery and allows the
prioritized delivery of packet, maintains the
multiple buffer delivery of packet Our
objective is to maximize weighted
Throughput, defined by the total value of the
packets sent.

1. Introduction Figure 1. The buffer management


inside a network switch.
Network switches inside routers of
the IP-based network infrastructure are Buffer size is considered to be
critical in implementing the Internet’s finite and drops the packet if overflow
Functionalities. Packets arrive at the occurs. This paper proposes the concept of
network switches in an online manner. The multiple buffers for managing large
buffer management policy in the network number of packets Using multiple buffer it
Switches is in charge of two tasks: packet able to handle large number of packets in
queuing and packet delivery. When new the network and also based on the priority
packets arrive, buffer management decides of the packet it will be delivered.
which ones to accept and queue for potential
delivery. Since the buffer has finite buffer
slots, it may not be able to accommodate all
2. Competitive FIFO Buffer Management In optimal offline algorithm each
for Weighted Packets time step, ROS greedily accepts packets
and drops the minimum-value one if
They consider efficient online packets overflow happens. Then ROS
algorithms for buffer management in sends a pending packet with the maximum
network switches. We study a FIFO value. Since those sent packets do not
buffering model, in which unit-length need to obey the FIFO order in the
packets arrive in an online manner and each delivery sequence, at the end of each time
packet is associated with a value (weight) step, all unsent packets (if any) will be
representing its priority. The order of the kept in the buffer.
packets being sent should comply with the
order of their arriving time. The buffer size 3. Competitive Queuing Policies for QoS
is finite. At most one packet can be sent in Switches
each time step. They[2] design competitive Packet scheduling in a network
online FIFO buffering algorithms, where providing differentiated services, where
competitive ratios are used to measure each packet is assigned a value. various
online algorithms’ performance against the queueing models for supporting QoS
worst-case scenarios. They first provide an (Quality of Service)[3]. In the
online algorithm with a constant competitive nonpreemptive model, packets accepted to
ratio 2. Then, They study the experimental the queue will be transmitted eventually
performance of their algorithm on real and cannot be dropped. The FIFO
Internet packet traces and compare it with preemptive model allows packets accepted
all other known FIFO online competitive to the queue to be preempted (dropped)
algorithms. They provide an online prior to their departure, while ensuring that
algorithm ON with a constant competitive transmitted packets are sent in the order of
ratio. They experimentally evaluate arrival. In the bounded delay model,
algorithm ON and all known competitive packets must be transmitted before a
buffer management algorithms. certain deadline, otherwise it is lost (while
transmission ordering is allowed to be
arbitrary). In all models the goal of the
2.1. A Competitive Online Algorithm ON buffer policy is to maximize the total value
of the accepted packets. Let a be the ratio
Fei Li[2] describes an online algorithm between the maximal and minimal value.
called ROS (ROS stands for Relaxed Online For the non-preemptive model They derive
Optimal Solution). ROS Works in a relaxed a O(log a) competitive ratio, both
model in which the FIFO order constraint over exhibiting a buffer policy and a general
sending packets is not demanded. A good Lower bound.
characteristic of ROS results that can per
mutate the packet sending sequence of ROS to 4. Competitive Queue Policies for
get an optimal offline algorithm OPT for the Differentiated Services
FIFO buffering model. ROS is used to
calculate ON’s competitive ratio. In Competitive Queue Policies for
Differentiated Services the packets are
2.2. An optimal offline algorithm OPT tagged as either being high or low priority
packets. Outgoing links in the network are
serviced by a single FIFO queue. This
model [4] gives a benefit of α>= 1 to each Simulations of a TCP/IP network are used
high priority packet and a benefit of 1 to to illustrate the performance of RED
each low priority packet. A queue policy gateways. Random Early Detection
control which of the arriving packets are gateways are ineffective mechanism for
dropped and which enter the queue. Once a congestion avoidance at the gateway, in
packet enters the queue it is eventually sent. cooperation with network transport
The aim of a queue policy is to maximize protocols. If RED gateways packets when
the sum of the benefits of all the packets it the average queue size exceeds the
delivers. W. Aiello, Y. Mansour, S. maximum threshold, rather than simply
Rajagopolan, and A. Rosen analyze and setting a bit in packet headers, then RED
compare different queue policies for this gateways control the calculated average
problem using the competitive analysis queue size. This action provides an upper
approach, where the benefit of the once bound on the average delay at the
policy is compared to the benefit of an gateway. The probability that the RED
optimal offline policy. They derive both gateway chooses a particular connection to
upper and lower bounds for the policies and notify during congestion is roughly
in most cases bounds are tight. proportional to that connection’s share of
the bandwidth at the gateway. This
5. Random Early Detection Gateways for approach avoids a bias against bursty
Congestion Avoidance traffic at the gateway.
Random Early Detection (RED)
gateways for congestion avoidance in 6. Proposed Work
packet-switched networks. The gateway In this paper we proposes the
detects incipient congestion by computing concept of implementing the multiple
the average queue size. The gateway could buffers for managing large number of
notify connections of congestion either by packets. In simple buffer not possible to
dropping packets arriving at the gateway or manage the more number of packets.
by setting a bit in packet headers. When the
average queue size exceeds a preset 6.1 Packet Analysis
threshold, the gateway drops or marks each Packet analysis, often referred to as
arriving packet with a certain probability, packet sniffing or protocol analysis,
where the exact probability is a function of describes the process of capturing and
the average queue size. RED gateways keep interpreting live data as it flows across a
the average queue size low while allowing network in order to better understand what
occasional bursts of packets in the queue. is happening on that network. Packet
During congestion, the probability that the analysis can help us understand network
gateway notifies a particular connection to characteristics, learn who is on a network,
reduce its window is roughly proportional to determine who or what is utilizing
that connection’s share of the bandwidth available bandwidth, identify peak
through the gateway. RED gateways are network usage times, identify possible
designed to accompany a transport-layer attacks or malicious activity, and find
congestion control protocol such as TCP. unsecured and bloated applications.
The RED gateway has no bias 6.1.1 Viewing Endpoints
against bursty traffic and avoids the global An endpoint is the place where
synchronization of many connections communication ends on a particular
decreasing their window at the same time. Protocol. For instance, there are two
endpoints in TCP/IP communication: the IP request separate buffer is maintained.
addresses of the systems sending and Buffer has a fixed number of slots. All the
receiving data, 192.168.1.25 and arriving packets are fed into the buffer and
192.168.1.30. An example on Layer 2 would delivered to the client requests.
be the communication taking Place between
two physical NICs and their MAC
addresses. The NICs sending and receiving
data have addresses of 00:ff:ac:ce:0b:de and
00:ff:ac:e0:dc:0f, Making those addresses
the endpoints of communication.

Fig 6.2 Multiple Buffer Management

6.3 Packet Priority


When a network segment becomes
congested, the hub-and-switch workload
results in the delay or dropping of packets.
On a network using packet-priority values,
Fig 6.1 Viewing end points a packet with a higher priority receives
preferential treatment and is serviced
before a packet with a lower priority.
N Source Destination Destination Capture Capture
Source IP
o MAC MAC IP d Time d Priority is given to the application
Length packets based on the type of data packets.
Thu
Nov 19 Each data packets have different priority
192.168.0.25
0 00:26:18:51: ff:ff:ff:ff:ff:ff 192.168.0.11
5
09:47:02 209 to the delivery. Priority is set based on the
3b:2f IST delay value of the arriving packets. Each
2009
data packet has a delay value. During the
Thu packets in the buffer the delay value of
Nov 19 packet is analyzed and based on the delay
00:26:18:51: 00:08:5c:8c:e 203.145.184.
1 192.168.0.11 09:47:02 52
3b:2f 7:06 32
IST constrain the packet delivered to the
2009 clients. Thus allows the prioritized
delivery of packet.

Table 6.Packet Analysis Conclusion


6.2 Multiple Buffer Management
Multiple Buffers are managed We conclude that for practical
between clients in order to maintain the usage of these priority algorithms,
large number of incoming packets. For each thorough experiments need to performed
and analyzed to pick the best algorithm in
real applications. The Multi-buffering model
and measure its performance experimentally
measured using Internet traces.

References

[1] The Internet traffic archive.


http://ita.ee.lbl.gov.

[2] Fei Li Competitive FIFO Buffer


Management for Weighted Packets .In
proceedings of the 7th Annual
communication and services Research
Conference,2009

[3] N. Andelman, Y. Mansour, and A. Zhu.


Competitive queuing polices for QoS
switches. In Proceedings of the 14th Annual
ACM-SIAM Symposium on Discrete
Algorithms (SODA), pages 761–770, 2003.

[4] W. Aiello, Y. Mansour, S. Rajagopolan,


and A. Rosen. Competitive queue policies
for differentiated services. In Proceedings of
the 19th Annual Joint Conference of the
IEEE Computer and Communications
Societies (INFOCOM), pages 431–440,
2000.

[5] S. Floyd and V. Jacobson. Random early


detection gateway for congestion avoidance.
IEEE/ACM Transactions on Networking,
pages 397–413, 1993.

Anda mungkin juga menyukai