Anda di halaman 1dari 8

2012 32nd International Conference on Distributed Computing Systems Workshops

A Delay-based Routing Protocol for Human-oriented Delay Tolerant Mobile Sensor


Network (DTMSN)

Hui Zhao∗ , Ming Liu∗


∗ School
of Computer Science and Engineering
University of Electronic Science and Technology of China
Email:jenniferzhao09@gmail.com,csmliu@uestc.edu.cn(corresponding author)

Abstract—With the vigorous development of IOT (Internet One typical application of DTMSN is human-oriented
of Things), sensor units carried by human beings will form the data gathering, in which the sensors collect the data which
largest DTMSN in the world. However, most of the existing are closely related to human body to achieve certain goal,
DTMSN routing protocols haven’t taken the characteristics
of human mobility into account, and may not work well in such as flu virus tracking to prevent the explosion of
human-oriented data gathering. In this paper, we analyze the devastating flu or air quality monitoring for tracking the
characteristics of human mobility, and propose a novel delay- average toxic gas taken by people everyday [1]. In these
based routing protocol (DRP) for human-oriented DTMSN. In applications, although samples can be collected at specific
DRP, a sensor node calculates the estimated data delivery delay locations, the most accurate and effective way is deploying
and takes it as a measure of the delivery capacity, the smaller
the estimated delay, the higher the delivery capacity of a sensor wearable sensing units at human body, which generates large
node. When two nodes meet, data messages are forwarded to scale human-oriented DTMSN. With the vigorous develop-
the one with smaller estimated delay. To minimize transmission ment of IOT (Internet of Things), sensor units wearing by
overhead, DRP employs the message rank and survival time human beings will form the largest DTMSN in the world.
to decide message’s transmission or dropping. We evaluate the Hence, it’s of great significance to design highly effective
proposed scheme in a real human moving scenario provided
by MIT Reality datasets, which collected the traces of 100 routing protocol for such networks. Among the existing
individuals of MIT over the course of 9 months. Simulation DTMSN routing protocols, most of them [4], [5], [6], [7]
results have shown that the proposed DRP routing protocol cannot adjust to the characteristics of DTMSN and work
achieves a better performance than some other DTMSN data poorly in data gathering. For example, due to the low data
gathering approaches. deliver ratio in direct transmission [4] and the tremendous
Keywords-DTMSN,data gathering,data delivery delay,real amount of energy consumption in epidemic algorithms [5],
human mobility traces these two basic routing schemes work poorly in practical
scenarios. Although mitigating the energy expense, MaxProp
I. I NTRODUCTION [6] and PREP [7], two variants of the epidemic protocol,
Data gathering is the most important function of wire- still have very high transmission overhead. RED [8] and
less sensor network (WSN). The traditional data gathering FAD [4] takes the characteristics of DTMSN into account
schemes usually rely on a large number of densely deployed and make routing decisions based on historic records. They
sensor nodes to form a well connected mesh network. achieve better routing performance compared with other
Sensors in the network collect the target data and jointly works, but their approach to predict delivery capacity is
transmit them to the sink nodes. These schemes, however, error-prone, especially when the source is far away from
may not work well in the scenarios with poor connectivity the sink; In addition, as they are not specially designed for
due to sparse network density and mobility of sensor nodes human-oriented DTMSN, they may not work well in human-
(e.g., air quality monitoring [1], and wildlife tracking [2]). In oriented scenario.
order to collect the data information in the aforementioned In this paper, we analyze the characteristic of human
scenarios effectively, Delay Tolerant Mobile Sensor Network mobility, and propose a novel delay-based routing protocol
(DTMSN)[3] was proposed. A DTMSN generally consists (DRP) for human-oriented DTMSN. The major contributions
of two types of nodes, namely mobile sensor nodes and of our work lie in the following:
sink nodes. The mobile sensor nodes which are intermittent
connected are attached to moving objects (such as human • We propose a new data gathering approach for human-
beings, animals, or vehicles) for data information gathering, oriented DTMSN, which estimates the data transmis-
while the sink nodes are either placed at special locations sion delay of a sensor node according to the history
or taken by mobile objects to collect data from sensors records, and replicates messages selectively to nodes
and forward them to the end user. As the connectivity is that have lower estimated data delivery delay;
intermittent, a certain delay is tolerable in such networks. • We introduce an effective queue management scheme.

1545-0678/12 $26.00 © 2012 IEEE 201


DOI 10.1109/ICDCSW.2012.108
According to this scheme, messages with too large utilization. In [4], the authors propose a FAD protocol
survival time should be dropped to make the full use to increase the data delivery ratio in DTMSN. Besides
of network bandwidth and to reduce network energy using the same delivery probability calculation method as
consumption; RED, FAD further discusses how the replication of the data
• We use a real human moving scenario provided by over the sensor network can be constrained using a fault
MIT Reality datasets [9] to simulate the proposed tolerance value associated to each data message. However,
protocol. Compared with the previous works which that protocol still has quite a high overhead.
assume that the nodes move randomly, our results are In recent years, social structures have been exploited to
more persuasive; help routing in intermittently connected networks. In [10],
The rest of the paper is organized as follows: Sec. 2 makes the authors utilize the communities formed by nodes in the
a brief overview of related work. Sec.3 presents the network network for controlled flooding. They assume that nodes
model. Sec. 4 introduces the proposed DRP. Sec. 5 shows mainly remain inside their community and sometimes visit
the effectiveness of DRP via simulation. Sec. 6 concludes others. To route a message to a destination, a node may
the paper. forward that message to a node that belongs to the same
community as the destination. Their work provides a theo-
II. R ELATED WORK retical hypothesis for community based routing, but there is
Routing approaches in delay tolerant scenario have been not yet any empirical evaluation. In [11], the authors present
explored by many research groups. In [4], the authors present BUBBLE. According to this protocol, data messages should
a simple routing protocol for DTMSN, where data is only be forwarded to the nodes which are more popular in the
allowed to be delivered when sensors are in direct proximity networks, as the authors believe that the popular nodes have
of the sinks. That protocol has very little communication more chances to deliver data successfully. The simulation
overhead but much long delivery delay. In [5], Vahdat et shows that this mechanism improves the performance of
al. propose an epidemic routing protocol to increase the routing in DTN obviously. In [12], Daly et al. present
data delivery ratio in partially connected networks. The the SimBet routing metric which is comprised of both a
communication overhead here is quite high since too many node’s centrality and its social similarity. If the destination
message copies are produced. node is unknown to the sending node or its contacts in
Apart from the most basic and simple routing schemes Simbet, the message is routed to a structurally more central
mentioned above, many other approaches have been pro- node where the potential of finding a suitable carrier is
posed to address the data gathering problem in DTMSN. dramatically increased. Simulation results demonstrate that
The ZebraNet project [2] proposes a history-based strategy Simbet routing comes close in terms of performance to that
for routing. The routing decision here is made according of epidemic protocol [5], without the additional overhead of
to a sensor node’s past success rate of transmitting data redundantly forwarding the message.
packets to the sink nodes directly. However, the data delivery
III. S YSTEM M ODEL
ratio is very low. Paper [4]also finds out that protocol fails
in delivering data messages generated by the sensor nodes Definition3.1: A geographical region formed by some
that are far away from the sink nodes. Other endeavors sensor nodes moving within this scope for most of the time
aiming to enhance the performance of DTMSN routing is called a community.
include paper [8], [4]. In [8], a replication based efficient We assume that one community consists of N number of
data delivery (RED) protocol is presented. RED consists of nodes and one sink node. A typical community is as shown
two components: data delivery and message management. in Fig.1, in which the red triangle represents the sink node
Data delivery uses a history-based method as ZebraNet to and the black circles represent the sensor nodes. Moreover,
calculate the delivery probabilities of sensor nodes. Con- we further assume some additional characteristics in our
cretely, let ξi denote the delivery probability of sensor i. model.
ξi is initialized with zero and increases upon an event of • Sensor nodes are carried by human beings, and move
message transmission while reducing with time expiration. with their carriers;
The message management algorithm decides the optimal • The sink node is located at a fixed place in its commu-
erasure coding parameters based on sensor’s current delivery nity, and has no energy restriction;
probability for improving the data delivery ratio. However, • All the sensor nodes are loosely synchronized;
the optimization of erasure coding parameters is usually The community we define corresponds to the organized
inaccurate because they are calculated according to the human group in our society. It’s worth to note that, node
current data delivery probability of source node, especially mobility model has a great impact on the performance of
when the source is very far away from the sinks. In addition, a routing protocol[13], [14]. Among the existing mobility
propagating many small messages in the network may incur model, the Random Mobility Model assumes that the nodes
further processing overhead and inefficiency of bandwidth move randomly, which obviously couldn’t reflect human

202
6% 5%

27%

52%

Community

Figure 1. network model

Figure 2. Visiting frequency of different places


moving characteristics; the community-based mobility mod-
el [15], [16], [17] tries to abstract the characteristics of
human movements, but without a widely accepted standard message periodically, and use the receiving delay of each
until now. In our network model, we do not constrain the hello message as an approximation of the data delivery
movements of nodes using any mobility model, and just delay of a sensor node at different time point. As the
assume that they move as in a real circumstance. links are bidirectional, this approximation is reasonable.
IV. D ELAY BASED ROUTING P ROTOCOL The broadcasted hello message includes only two fields:
ID and GT, representing the message ID and its generation
In human-oriented data gathering application, sensors are
time respectively. The nodes stays in direct proximity of
carried by human beings, and data forwarding chances are
sink first receive the hello message, and then broadcast this
brought by human contacts. Thus, we will first analyze
message in the whole community through contacts with
the characteristics of human mobility before presenting our
other nodes. Due to the intermittent connectivity of DTMSN,
routing protocol.
each node receives the same hello message with different
We notice that there is certain regularity and repeatability
delay. Whenever receiving a new hello message m (as is
in every person’s life. In [18], it is pointed out that 93%
determined by the ID), the node will calculate the current
of human mobility is predictable. The author did a survey
delivery delay as follows:
on the mobility of a phone user over the course of three
months, and the result is shown in Fig.2. The nodes represent Delay(i) = RT (m) − GT (m) (1)
the locations visited by the user, and links represent the
observed movements between these locations. The uneven Where GT (m) represents the generation time of m, and
node sizes, corresponding to the percentage of time the user RT (m) represents the receiving time of m by node i.
spent in the vicinity of the particular tower, indicate that Then, the history average delivery delay of node i is
individuals tend to spend most of their time in a few main updated according to the following formula:
locations. Furthermore, we observe in our life that during a Avgdelay(i) = αDelay(i) + (1 − α)Avgdelay(i) (2)
specified period of a day, one is usually attached to a certain
place, meeting almost the same people. For example, during Where α is a parameter which determines the percentage the
the working hours, one usually stays at the office, with current delivery delay holds in estimating the data delivery
his/her most contacts occur with colleagues. Considering delay of node i.
these characteristics, the data delivery capacity of a node (a As mentioned above, we use the history average delivery
person) could be predicted by the history records very well. delay to estimate the delivery delay at one moment, and
In this paper, we take the history average delivery delay as take it as the measure of delivery capacity, the smaller the
an approximation of the estimated data delivery delay and estimated delay, the higher the delivery capacity of a sensor
make routing decisions based on this value. The details on node.
routing protocol will be presented in the following part. 2) Data transmission algorithm: Now we consider a
sensor i , which has a total number of N messages in
A. Data Transmission the data queue ready for transmission and is moving in-
1) Data delivery delay estimation: Data transmission to
∑ the communication range of a set of Z ′ sensors. Let
decision is made based on the estimated data delivery delay = Ψz |1 ≤ z ≤ Z represent the Z ′ nodes. Sensor i first

of each sensor node at the moment, which indicates how learns their current average transmission delay via simple
soon a sensor node could deliver a data message to the sink. handshaking messages and then replicates all the messages
Obviously, it’s difficult for a sensor node to get the data in its queue to a subset of the Z ′ sensors, which have lower
delivery delay without receiving acknowledgement from the estimated delivery delay than that of node i. The pseudo-
sink node. Therefore, we let the sink node broadcast a hello code of the data transmission algorithm is shown as follows:

203
Algorithm1.Message forwarding process
1: ϕ = 0
its own data queue again. Among these messages, we let
2: for z = 1; z <= Z ′ ; z + + do // identify receiving sensors message of sort a have the highest rank, sort b medium, and
3: if Avgdelay(i) > Avgdelay(ψz ) then sort c the lowest rank. That is because messages of sort a
4: ϕ=ϕ ∪ ψz
5: end if
haven’t been transmitted, while that of sort b and sort c have
6: end for been transmitted, with a higher chance of being delivered to
7: for n = 1; n <= |ϕ|; n + + do // |ϕ| represents the number the sink node.
of nodes in ϕ
8: forwardmessage ( i, ϕn ) //ϕn for all n ∈ [1, |ϕ|] Let Ci denotes the clock of sensor i and ξim be the survival
9: end for time of message m in the queue of sensor i. Recall that
in our network model, all the sensor nodes have loosely
synchronized clocks, thus, our strategy for determining a
The data delivery process is shown in Fig.3. Suppose message’s survival time is as follows: when a message is
that S1 → S2 → S3 → S4 → · · · → Sn is the path generated, its survival time is initialized to be zero. Let’s
along which the data messages are successfully forwarded consider a sensor n, which is transmitting a data message
to the sink, where Si represents a sensor node carried by one m to a nearby sensor node i. Since the propagation time
person. If we assume that the persons meeting each other between two adjacent nodes with short distance could be
have a certain social relationship, then this path can be used ignored, thus the initial value of ξim remains the same as
to represent a social relationship chain. In this chain, Sn ξnm . If a source message has transmitted to its next hop and
could communicate with the sink node directly, and it has it is inserted into node’s queue again, its survival time is
the lowest data delivery delay; while S1 is the furthest one also assumed to be equal to the value before transmitting.
with respect to sink node, thus have the longest data delivery Furthermore, each sensor maintains a timer. Once the timer
delay. As is analyzed above, this social relationship changes timeouts, the message’s survival time increases. The general
slowly, and this may also explain why we could use the operations are presented as Algorithm 2, where parameter
history average delivery delay to estimate the delivery delay K denotes the maximum queue size of sensor.
in the future.
Algorithm 2.Survival time updating
1: a[i] ← Ci ; // a[i] keep the current value of Ci
2: if receiving a new generated message m from node i
Sn
S3 S4 3: ξim =0
Sink 4: else
S2
5: if receiving a message m form adjacent node n
6: ξim =ξnm
S1 7: else
8: if receiving a transmitted message m from node i
9: ξim =ξim
10: endif
11: For( m = 1; m <= K; m + +) // m represent messages
Figure 3. Data delivery process maintained by sensor i
12: while m is in the queue of node i
13: ξim =ξim + +
14: endwhile
B. Queue management 15: endfor
1) Message’s rank & survival time: In opportunistic net-
works like DTMSN, multiple copies of the same messages 2) The implementation of queue management scheme:
would be generated and buffered by different sensors, which The queue management scheme is based on both the rank
results in redundancy. The goal of the queue management and the survival time of the message. We believe that
scheme is to determine message’s transmission order in message of higher rank is more important and should be
the buffer queue and to determine which message should forwarded with a higher priority. This is done by arranging
be dropped when the queue is full. The main idea of our the messages in the queue with a decreasing order of their
queue management scheme is to employ the rank as well as rank. Furthermore, for messages with the same rank, priority
survival time to signify the importance of a given message. should be given to those messages that have smaller survival
Each sensor has a data queue that contains data messages time. A message will be dropped in the following two
ready for transmission. The data messages of a sensor come occasions. First, if the queue is full when a message arrives,
from three sources. (a) After the sensor acquires data from its rank is compared with the message at the end of the
its sensing unit, it creates a data message, which is inserted queue. If the rank is equal, but the new message has a larger
into its data queue; (b) When the sensor receives a data survival time, it is dropped. If the rank is equal, but the new
message from other sensors, it inserts the message into its message has a smaller survival time, the message at the end
data queue; (c) After the sensor sends out a data message to of the queue is dropped, and the new message is inserted
a non-sink sensor node, it may also insert the message into into the queue at appropriate position according to its rank

204
and survival time. If the rank of the new message is higher,
the message at the end of the queue is dropped, and the
new message is also inserted into the queue at appropriate
position. Second, once the survival time of a message in
the process of updating exceeds the network’s delay tolerant
threshold, the message is dropped. This is to reduce network
energy consumption, given that the message either has been
delivered to the sink node with a high probability by other
sensors or has been invalid in our application.

V. S IMULATION Figure 4. Friendship network in Reality Mining project


We simulate and evaluate four protocols, including the
Table I
proposed DRP, FAD, direct transmission and epidemic in S IMULATION RESULTS WITH DEFAULT PARAMETERS
a real human moving scenario provided by MIT Reality
datasets. FAD is one of the most influential DTMSN routing Parameter Default value
protocols, while direct transmission and epidemic protocol broadcast interval(minute) 6
Maximum queue size of sensor(message) 300
actually provide us with a performance baseline. We will Size of data message(byte) 200
first introduce the datasets we use, and then compare the Size of hello message(byte) 10
performance among the four protocols and evaluate the Message generation rate (message/minute) 0.05
Network tolerant delay(minute) 1000
impact of different parameters.
Initial energy of sensor node (joule) 10
α of DRP 0.6
A. Datasets introduction
Reality datasets collected the traces of 100 individuals
from MIT over the course of 9 months by letting each of We compare the performance of the four protocols under
them carry a Bluetooth-enabled mobile phone. This phone the default parameters, and the results are presented in
can record and timestamp the Bluetooth IDs encountered Table 2. As we can see, the DRP protocol achieves a
as well as the cell tower ID providing services for it. As higher data delivery ratio (90.7%) at relatively lower delivery
the phones are carried by people, an encounter means a delay with nearly the same overhead as FAD. The direct
communication opportunity in DTMSN. Moreover, we use transmission protocol causes little overhead, but the delivery
the static Bluetooth device frequently encountered by other ratio is extremely low. The epidemic protocol has the highest
Bluetooth devices to represent the sink node in DTMSN. overhead, however, the data delivery ratio is not as high as
Fig.4 is the friendship network generated through a survey it should be.
of the 100 individuals, which is clearly divided into two For DRP, it takes the estimation of data delivery delay
cliques. This may be explained by the fact that: among the as a measure of delivery capacity, and data messages are
100 individuals, seventy-five are either students or faculty forwarded to the ones with lower estimated delay. To some
in the MIT Media Laboratory, while the remaining twenty- extent, it could guarantee that the data is delivered towards
five are incoming students at the MIT Sloan business school the sink node step by step. Moreover, DRP employs the
adjacent to the Laboratory[4]. According to our network message rank and survival time to decide message’s trans-
model, one sink node is placed in one community for data mission or dropping. The using of message ranks make sure
gathering, so we just consider the MIT Media Laboratory that new generated messages have more delivery chances;
community in our simulation, which includes 75 people. The The message survival time could eliminate the old enough
sink node is chosen to be the static device encountered most messages, which may have already been delivered to the
frequently by these people. sink node, but still exhausts the network resources. For FAD,
the delivery probability of a sensor node is initialized with
B. Performance comparison zero and increases upon an event of message transmission
The performance metrics we use in our simulation are: while reducing with time expiration. Due to the sparseness
data delivery ratio, data delivery delay, and data delivery and intermittent connectivity of DTMSN, the frequency of
overhead (represented by the average copies for each mes- encountering other sensor nodes or the sink node is very low,
sage). The simulation parameters as well as their default thus, the transmission probability given by FAD could not
values are shown in Table 1. Moreover, as the performance reflect the delivery capacity of a node very well. Regarding
of FAD is influenced by the timer expiration value ∆ and Direct transmission, the sensor nodes can only transmit mes-
parameter α greatly, we adjust these two parameters to get sages to the sink node directly, for sensors never reachable
the close-to-optimal performance for fair comparison. to the sink node, their data messages will be lost forever,

205
thus the delivery ratio is very low. For epidemic protocol, delivery ratio, the survival time of data messages mustn’t be
it produces a large amount of copies, which exhausts the too small. In our experiments, we set the network tolerant
buffering resources rapidly, causing serious data dropping delay as 1000 minutes, and the data delivery ratio of DRP
phenomenon and low delivery ratio. is nearly 90%.

Table II D. Impact of broadcast interval


S IMULATION RESULTS WITH DEFAULT PARAMETERS
Fig.6 illustrates how the broadcast interval of hello mes-
Protocol DRP FAD DT Epedemic sage at the sink node affects our proposed protocol. As
Delivery ratio(%) 90.7 87.5 46.4 76.1 shown in Fig.6, while the broadcast interval varies from 1
Average copies 6.1 6.2 1.0 9.3
Average delay(s) 525 623.1 993.4 405.2
minute to 6 minutes, the data delivery ratio and delivery
delay of DRP have seen little changes, and the delivery
overhead decreases slightly with this varying. This is be-
C. Impact of network tolerant delay cause the activities of human beings are stable, which means
that the delivery capacity of the sensor nodes carried by
This group of experiment illustrates the impact of the
human beings changes slowly. When the broadcast interval
network tolerant delay on the data delivery ratio of the four
is too small, some randomness might be brought and affect
protocols. The network tolerant delay indicates the survival
the accuracy of the judgment on delivery capacity of a
time of each message, and is represented by TTL (Time To
node, which thus incurs more delivery overhead. Moreover,
Live) here. As shown in Fig.5, while varying TTL from 200
a smaller broadcast interval of hello message will lead to
100
more extra overhead. Hence,we set the broadcast interval as
90
6 minutes in our simulation.
80

70
E. Impact of queue sizes
Data delivery ratio

60 This group of experiment illustrates the impact of the


50
queue sizes on the performance of the four protocols. As
40
shown in Fig.7(a), with the growing of the queue sizes, the
30
DRP
20 FAD
data delivery ratio of the four protocols are all improved,
10
Epidemic this is because larger queue sizes could accommodate more
DT
0 data messages, and the data messages have more chances
200 400 600 800 1000
TTL to be delivered to the sink node before they are dropped.
As seen from Fig.7(b), the average message copies of the
Figure 5. Impact of network tolerant delay four protocols increases with the growing of the queue sizes,
this is also because large queue sizes could accommodate
minutes to 1000 minutes, the data delivery ratios of the four more data messages. We find that DRP could control its
protocols are all improved. This is because longer survival average message copies effectively. As shown in Fig.7(c),
time means more chances to be delivered to the sink node with the increasing of the queue sizes, the data delivery
before being dropped. While the TTL is among 200 minutes delay of the four protocols all increases. This is because
to 400 minutes, the data delivery ratios of the four protocols when the queue sizes increases, those messages with high
are all very low, and epidemic outperforms the other three delivery delay have more chances to be delivered to the
protocols. This is because small TTL leads to frequent sink node. Overall, at different buffer sizes, DRP outperform
message dropping and a great amount of messages are direct transmission and epidemic obviously, and outperform
dropped before being delivered to the sink node. In this case, FAD in both delivery ratio and delivery delay.
the queue sizes are adequate and epidemic could make every
chance to deliver data messages, thus it achieves the highest F. Network life analyze
delivery ratio. However, as the TTL continues increasing, This group of experiment shows the lifetime of the
the number of survival messages increases obviously, and network using the four protocols. We assume that the energy
the queue sizes become a scarce resource. As a result, the needed in each transmission and receiving is as specified in
epidemic protocol leads to serious buffer overflow, and its [19], and the results are presented in Table 3. We assume the
data delivery ratio becomes lower than DRP and FAD. In network dies when half of the sensor nodes exhaust their en-
direct transmission, a sensor node could only transmit while ergy. We can see that the direct transmission protocol enjoys
meeting the sink node, thus it has the lowest data delivery the longest network lifetime, since sensor doesn’t receive
ratio all the time. As the nodes in the datasets we use in and transmit any other messages except those generated by
our simulation are very sparse, it may take one node a the sensor itself, and thus much energy can be saved. The
long period to meet another node. To achieve an ideal data epidemic protocol maintains the shortest network lifetime.

206
100 10 600

80 8 500

Data delivery ratio(%)

Average copies

Average delay
60 6 400

40 4 300

20 2 200

0 0 100
1 2 3 4 5 6 1 2 3 4 5 6 1 2 3 4 5 6

Broadcast interval Broadcast interval Broadcast interval

(a) data delivery ratio (b) average copies (c) transmission delay

Figure 6. Impact of broadcast interval

100 15 1200

80 12 1000
DRP
Data delivery ratio(%)

FAD
Average copies

Average delay
60 9 800 Epidemic
DT

40 6 600
DRP
DRP FAD
FAD Epidemic
20 3 400
Epedemic DT
DT
0 0 200
100 200 300 400 500 600 100 200 300 400 500 600 100 200 300 400 500 600
Queue size Queue size Queue size

(a) data delivery ratio (b) average copies (c) transmission delay

Figure 7. Impact of queue sizes

This is because too many messages copies exhaust much minimize transmission overhead, we introduce an effective
sensor energy in flooding. Moreover, we also obtain that queue management scheme, which utilizes the message rank
DRP attains longer network life than FAD, demonstrating and survival time to determine message’s transmission or
the effectiveness of the proposed strategy. Though extra dropping. Finally, we evaluate the proposed scheme in a
energy is needed in DRP to disseminate the hello messages real human moving scenario, and the results show that the
broadcasted by the sink node, the total energy consumption proposed DRP achieves a higher delivery ratio at lower
of DRP is a little less than FAD. This is because the size of delivery delay with reasonable overhead.
the hello message is very small and the broadcasting interval
does not need to be very short due to the stable human R EFERENCES
activity.
[1] Y. Wang, H. Wu, H. Dang, and F. Lin, “Analytic, simulation,
Table III and empirical evaluation of delay/fault-tolerant mobile sensor
N ETWORK LIFE WITH DEFAULT PARAMETERS
networks,” IEEE Transactions on Wireless Communications,
2007, 1(11): 3287-3296.
Protocol DRP FAD DT Epedemic
Lifetime 10.71 10.28 7.48 65.6
[2] J. Philo, O. Hidekazu, and W. Yong, “Energy-efficient com-
puting for wildlife tracking: design tradeoffs and early expe-
VI. C ONCLUSION riences with zebranet,” ACM Operating System Review, 2002,
36(5):96-107.
This paper focuses on the data gathering in human-
oriented DTMSN. After analyzing the unique characteristics [3] Y. Wang, F. Lin, and H. Wu, “Poster: efficient data trans-
of human mobility, we propose a delay-based routing pro- mission in delay fault tolerant mobile sensor networks (dft-
tocol (DRP) for human-oriented DTMSN. In DRP, the sink msn),” In: Proceedings of IEEE International Conference on
Network Protocols (ICNP’05), 2005.
node broadcasts a hello message periodically; each sensor
node calculates the estimated data delivery delay based on
[4] Y. Wang and H. Wu, “Delay/fault-tolerant mobile sensor
the received hello message and takes it as a measure of network (dft-msn): a new paradigm for pervasive informa-
the delivery capacity. When two nodes meet, data messages tion gathering,” IEEE Transactions on Mobile Computing,
are forwarded to the one with smaller estimated delay. To 2005,6(8):1021-1034.

207
[5] A. Vahdat and D. Becker, “Epidemic routing for partially con-
nected ad hoc networks,” Technical Report Vol.CS-200006:
Durham, 2000.

[6] J. Burgess, B. Gallagher, D. Jensen, and B. Levine, “Maxprop:


Routing for vehicle-based disruption-tolerant networks,” In:
Proceedings of IEEE INFOCOM, 2006.

[7] R. Ramanathan, R. Hansen, P. Basu, R. Hain, and R. Krish-


nan, “Prioritized epidemic routing for opportunistic network-
s,” In Proceedings of ACM MobiOpp’07, 2007.

[8] Y. Wang and H. Wu, “Replication-based efficient data delivery


scheme (red) for delay/fault-tolerant mobile sensor network
(dft-msn),” In: Enrico Gregori, eds. Proc. Of Fourth Annual
IEEE International Conference on Pervasive Computing and
Communications Workshops, 2006, 485-489.

[9] N. Eagle and A. Pentland, “Reality mining: sensing


complex social systems,” Personal Ubiquitous Computing,
2006,10(4):255-268.

[10] A. Lindgren, A. Doria, and O. Schelen, “Probabilistic routing


in intermittently connected networks,” In Proc. SAPIR, 2004.

[11] H. Pan and C. Jon, “Bubble rap: Social-based forwarding in


delay tolerant networks,” MOBIHOC, 2008.

[12] E. Daly and M. Haahr, “Social network analysis for routing in


disconnected delay tolerant manets,” In Proc. SAPIR, 2008.

[13] T. Camp, J. Boleng, and V. Davies, “A survey of mobility


models for ad hoc network research,” Wireless Communica-
tions and Mobile Computing (WCMC): Special Issue on Mo-
bile Ad Hoc Networking: Research, Trends, and Applications,
2001, 2(5):483-502.

[14] F. Bai, N. Sadagopan, and A. Helmy, “Important: a framework


to systematically analyze the impact of mobility on perfor-
mance of routing protocols for adhoc networks,” INFOCOM,
2003,825-835.

[15] Q. Zheng, X. Hong, and J. Liu, “An agenda based mobility


model,” In Proc of IEEE ANSS, 2006.

[16] M. Musolesi and C. Mascolo, “In: Proc. of the 2nd int’l


workshop on multi-hop ad hoc networks: From theory to
reality,” In Proc of IEEE ANSS, 2006, 31-38.

[17] W. Hsu, T. Spyropoulos, and e. a. K. Psounis, “Modeling


time-variant user mobility in wireless mobile networks,” In:
Proc. of the 26th IEEE Int’l Conf. on Computer Communi-
cations, 2007,758-766.

[18] C. Song, Z. Qu, N. Blumm, and A. Barabasi,


“Limits of predictability in human mobility,” Science,
2010,327(5968):1018-1021.

[19] W. Heinzelman, A. Chandrakasan, and H. ., “An application-


specific protocol architecture for wireless microsensor net-
works,” IEEE Trans. Wirel. Commun, 2002, 1, 660-670.

208

Anda mungkin juga menyukai