366 IEEE TRANSACTIONS ON MOBILE COMPUTING, VOL. 2, NO.

4, OCTOBER-DECEMBER 2003

On Route Lifetime
in Multihop Mobile Ad Hoc Networks
Yu-Chee Tseng, Senior Member, IEEE, Yueh-Feng Li, and Yu-Chia Chang

Abstract—One wireless network architecture that has received a lot of attention recently is the mobile ad hoc network (MANET). It is
attractive because the network can be quickly deployed without the infrastructure of base stations. One main feature of MANET is that
mobile hosts may communicate with each other through a sequence of wireless links (i.e., in a multihop manner). While many routing
protocols have been proposed for MANET by considering criteria such as length, quality, bandwidth, and signal strength [1], [8], [4], [5],
[7], [15], the issue of route lifetime has not been addressed formally. This paper presents a formal model to predict the lifetime of a
routing path based on the random walk model. Route lifetime is derived based on a probabilistic model. Through such investigation, we
hope to provide further insight into issues such as route selection, route maintenance, and network scalability related to MANETs.

Index Terms—Ad hoc networks, mobile computing, mobile networks, routing, wireless communication.

1 INTRODUCTION

T HEadvancement in wireless communications and light-

weight, small-size, portable computing devices have
made pervasive and mobile computing possible. One
may be less costly than a proactive protocol when host
mobility is high. Representative reactive protocols include
DSR (dynamic source routing) [8], ZRP (zone routing
wireless network architecture that has attracted a lot of protocol) [5], CBR [7], and AODV (Ad Hoc On Demand
attention recently is the mobile ad hoc network (MANET). A Distance Vector). A review of unicast routing protocols for
MANET is one consisting of a set of mobile hosts which MANET is in [16].
may communicate with one another and roam around at When choosing a routing path among several candi-
their will. Mobile hosts may communicate with each other dates, there are usually many factors to be considered, such
indirectly through a sequence of wireless links without as route length, route quality, signal strength, path
passing base stations (i.e., in a multihop manner). This bandwidth, and route lifetime. In this paper, we focus on
requires each mobile host serve as a router. A scenario of developing a formal model to evaluate the lifetime of a
MANET is illustrated in Fig 1.
routing path in a MANET. The result may be used in many
Applications of MANETs occur in situations like battle-
applications. For example, it can be used in the route
fields, festival grounds, outdoor activities, and emergency
discovery process to choose a most reliable path. It can be
rescue actions, where networks need to be deployed
used to determine when a route is likely to expire so that a
immediately, but base stations or fixed network infrastruc-
backup route can be searched in advance. As a longer route
tures are not available. For example, in an earthquake
is likely to suffer higher breakage probability, our result
disaster, all base stations may be down since there is not
may also be used to determine the cost-effectiveness of
electricity. In this case, a MANET driven by battery power
can be quickly deployed to set up a network environment. establishing a long route. Finally, the result may be used to
Such technology has been recently applied to wireless determine the proper size of a MANET considering the
sensor networks and personal-area networks too. route reliability versus route length trade off.
Extensive efforts have been devoted to the routing issue The route reliability issue has been addressed in several
on MANET. Routing protocols can be classified as proactive works. In the ABR routing protocol [18], the association
and reactive. A proactive protocol constantly updates the stability of links is accounted when choosing routing paths.
routing table of each host so as to maintain a (close to) The Signal Strength Adaptive (SSA) protocol [4] further
global view on the network topology. One representative considers the signal strengths in choosing routes. In [1], a
proactive protocol is the DSDV (destination-sequenced parameter called “affinity” is defined to characterize the
distance-vector) protocol [15]. On the contrary, a reactive signal strength and stability of a link. While choosing
protocol searches for a path in an on-demand manner. This routes, the “affinity” of a routing path is set to the minimum
link affinity in the path. The affinity concept is further
integrated with the transportation layer (i.e., TCP) by
. Y.-C. Tseng is with the Department of Computer Science and Information
accounting for the throughput of a routing path in [14]. A
Engineering, National Chiao-Tung University, Hsin-Chu, 30050, Taiwan.
E-mail: [email protected]. direction-prediction method is proposed in [17], where the
. Y.-F. Li and Y.-C. Chang are with the Department of Computer Science lifetime of a link is defined based on the current locations,
and Information Engineering, National Central University, Chung-Li,
roaming velocities, and roaming directions of the two
32054, Taiwan. E-mail: {wivern, jimchang}@axp1.csie.ncu.edu.tw.
neighboring hosts. Then, the lifetime of a routing path is
Manuscript received 30 Mar. 2002; revised 9 Apr. 2003; accepted 8 July 2003.
For information on obtaining reprints of this article, please send e-mail to: defined to be the minimum lifetime of each link in the path.
[email protected], and reference IEEECS Log Number 11-032002. The similar concept is applied to multicasting in [9].
1536-1233/03/$17.00 ß 2003 IEEE Published by the IEEE CS, CASS, ComSoc, IES, & SPS
TSENG ET AL.: ON ROUTE LIFETIME IN MULTIHOP MOBILE AD HOC NETWORKS 367

Fig. 1. An example of mobile ad hoc network (MANET).

In this paper, we present a formal model to predict the

lifetime of a route in a MANET. We assume that each
mobile host roams around following the random walk Fig. 2. Reliability of routing paths in a MANET.
model. Given a sequence of mobile hosts which form a
routing path, the joint probability distribution of route
2 SYSTEM MODEL
lifetime is derived based on the random walk model. This
differs from most existing works which directly calculate To motivate our work, let’s observe the example in Fig. 2.
the lifetime of a wireless link based on the current locations Suppose that we need to establish a route from source S to
and roaming directions of two neighboring hosts by destination D. In the network, two possible routes are
assuming that their roaming directions do not change. available: S ! A ! D and S ! B ! C ! D, where the
Also, while most works simply take the minimal lifetime of circles indicate radio coverage. The former route is shorter
each link in a path as the lifetime of the path, we formally in hop count, but is less reliable because host A can easily
derive the probability distribution. Based on our analysis, roam out of S’s and D’s radio coverage. The latter is longer,
extensive numerical and simulation results are presented. but might be more reliable since each intermediate host has
We comment that the purpose of this paper is not to a larger roaming area before the route will become broken.
present a new routing protocol for ad hoc networks. The above example has indicated a dilemma in route
Instead, our goal is to provide a formal model for selection. Most routing protocols tend to pick shorter routes
evaluating the lifetime of a given routing path. This can for efficiency in using wireless bandwidth. However, such
assist routing protocols in choosing from a multitude of routes may suffer from a higher chance of route breakage.
paths. Those routing protocols that are derived based on So, route reliability and route length are typically contra-
geographic forwarding (e.g., [10], [11]) are not applicable to dicting factors. This has motivated us to develop a formal
this case since no route needs to be established prior to model to predict the lifetime of routing paths in MANETs.
sending packets. We are aware that random walk is still too To develop a formal model, one has to adopt a roaming
simple to characterize human’s real mobility pattern, but is model for mobile hosts. In this paper, we use the discrete-
indeed one that is mathematically feasible to conduct time, random walk model, which has been widely used in
analysis. We expect, through this work, to motivate more personal communication services (PCS) networks [2], [3],
work in studying the impact of (more complicated) mobility [6]. Specifically, the area covered by the MANET is
patterns on route lifetime. The work in [12] assumes a partitioned into a number of hexagonal cells each of
random way-point model. Each host’s movement consists of a radius r, as shown in Fig. 3 (note, however, that, unlike
sequence of random length intervals called mobility epoch, PCS networks, there is no notion of base stations here). Each
during which the host moves in a constant direction at a cell is assigned a coordinate ðx; yÞ. There are two axes, one
constant speed for an exponentially distributed time pointing to the northeast and the other to the north. Sitting
interval. The speed and direction of each host vary in the center is cell ð0; 0Þ. The ith cell along the first axis is
randomly from epoch to epoch. Approximations of link sequentially numbered ði; 0Þ, while the jth cell along the
availability and path availability are derived, where link second axis is numbered ð0; jÞ. The coordinates of other
availability is defined to be the probability that a link is cells can be obtained by mapping onto these two axes, as
connected at time t þ
t given that it is connected at time t, normally done in Euclidean coordinates.
where
t is a time interval. However, the measure of link Following the formulation in most works, we further
availability does not exclude the possibility that the link partition cells into layers. Cell ð0; 0Þ is said to be on layer 0.
becomes disconnected at any instant during the time The six cells surrounding cell ð0; 0Þ are said to be on layer 1.
½t; t þ
t
. In comparison, in this work, we adopt a cellular Recursively, the outer cells surrounding cells at layer i are
model and we do consider such “cutoff” possibility in our said to be on layer i þ 1. The number of cells inside the nth
work. Under the same random way-point model, [13] layer (including the nth layer) is 3n2 þ 3n þ 1. In this paper,
further develops an efficient proximity model to measure the the transmission range of a mobile host will be modeled by
future stability of a given link. the number of layers that it can reach, assuming without
The rest of this paper is organized as follows: The system loss of generality that it is currently resident in cell ð0; 0Þ.
model is presented in Section 2. The probability distribution Although hosts can roam around in the real domain, we
of a route’s lifetime is derived in Section 3. Section 4 will work in a discrete domain by using cells as the basic
demonstrates some numerical and experimental results. units to model the locations of mobile hosts. Thus, the
Finally, Section 5 concludes this paper. smaller the cells are, the finer the locations are. Mobile hosts
368 IEEE TRANSACTIONS ON MOBILE COMPUTING, VOL. 2, NO. 4, OCTOBER-DECEMBER 2003

example, Fig. 4 shows three possible ways a link hx; yi

changes states. If the two hosts move along the arrows
marked by M1 , the new state is hx þ 1; yi. If they move along
M2 , the resulting state is the same. But, if they move along
M3 , the new state is hx  2; yi.
As mentioned earlier, a routing path will be modeled by
a sequence of vectors. After each time unit, each of its links
will change state according to the probability distribution in
Table 1. Accordingly, the state of the routing path will
change, too. If any link exceeds the maximum transmission
distance, the path is regarded as broken. Our goal is to
model the probability distribution of the route lifetime.
The first step is to do a state reduction to alleviate the
computational costs. The number of states for a wireless
link in an n-layers cellular network can be as large as
3n2 þ 3n þ 1. This number will increase rapidly as n
enlarges. To reduce the computational cost, we adopt the
model in [19] to merge equivalent states. Specifically, we
can partition an n-layer cellular network evenly into
12 equal-size sectors (refer to Fig. 5). Cells at neighboring
Fig. 3. The cellular system to model the locations of mobile hosts.
sectors with a reflective relation (with respect to the 12 axes)
are equivalent and can be merged into the same state. Thus,
the 12 sectors can be merged into one sector. The numbers
roam around in a cell-to-cell basis. We adopt a discrete-time
in Fig. 5a denote cell types after the merging, where cells of
model by dividing time into fixed-length units. Given any
the same number are of the same types. Fig. 5b shows the
cell which represents the current location of a mobile host,
sector containing all cell types after merging a 7-layer
the host will roam into the six neighboring cells in the next
network. The number of states is reduced from 169 to 20.
time unit with equal probabilities (1/6). This is what we
Formal derivation can be found in [19]. The basic observa-
mean by discrete-time, random-walk model.
tion is that all cells at the same distance from the central cell
along the 30-, 90-, 150-, 210-, 270-, and 330-degree axes will
3 ROUTE LIFETIME PREDICTION have the same probability distribution. This divides the
This section develops a model to predict the lifetime of a cellular network into six equivalent sectors. Further moving
routing path. Given two neighboring mobile hosts located from boundaries of each sector toward its central part, we
in cells ðx; yÞ and ðx0 ; y0 Þ, respectively, we denote the see, based on symmetry property, that the sector can be
wireless link from the former to the latter by a vector folded by half, thus further reducing the number of states
hx0  x; y0  yi. Thus, a routing path can be regarded as a by half. Depending of the value of n, the new number of
sequence of vectors, each representing one wireless link. For states becomes:
example, a routing path connecting hosts in cells ð0; 1Þ, 8
>
> 1 n¼0
ð3; 1Þ, and ð7; 3Þ can be written as ½h3; 0i; h4; 4i
(here, we < ðnþ1Þðnþ3Þ
4 n > 0 and n is odd
use brackets to denote a sequence). CðnÞ ¼
>
>
The vector representing a wireless link is called its state. : nðnþ4Þ
4 þ 1 n > 0 and n is even:
Next, we use the random walk model to formulate how a
wireless link changes states. Consider any wireless link Based on the state reduction, next we model the state
hx; yi connecting two hosts. After one time unit, each of the transition of a wireless link. Let n be the number of layers
two hosts may roam into one of its six neighbors. Thus, equal to a host’s radio coverage. Observe that a state vector
there are 36 combinations for the next state, each with the of length n may become n þ 2 in the next time unit (refer to
same probability of 1/36. Let the resulting vector be hx0 ; y0 i. Table 1). So, we need to consider an ðn þ 2Þ-layer network,
As some of 36 combinations will result in the same vector, which has Cðn þ 2Þ types of cells according to the above
there are only 19 possible hx0 ; y0 i. These vectors, together reduction. For each state, we can develop its state transition
with the associated probabilities, are shown in Table 1. For probability according to Table 1. For example, Fig. 6 shows

TABLE 1
The Probability Distribution for a Wireless Link to Switch from State hx; yi to State hx0 ; y0 i after One Time Unit
TSENG ET AL.: ON ROUTE LIFETIME IN MULTIHOP MOBILE AD HOC NETWORKS 369

Fig. 4. Example of link state changes.

Fig. 5. Merging equivalent states of wireless links: (a) partitioning the cells into 12 sectors and (b) equivalent states after merging.


the state transition of a wireless link when n ¼ 5. Note that P1 ði; tÞ if t ¼ 1
P2 ði; tÞ ¼
states h6; 0i, h5; 1i, h4; 2i, h3; 3i, h7; 0i, h6; 1i, h5; 2i, and h4; 3i P1 ði; tÞ  P1 ði; t  1Þ if t > 1:
are “absorbing” states since, once entering these states, the The above derivation is for one link. Next, we consider a
link is broken (hence, there is no exit).
routing path R, which consists of a sequence of k wireless
The state transition diagram of a wireless link can be
translated to a state transition matrix M such that each links ½i1 ; i2 ; . . . ; ik
at time 0. To simplify the derivation, we
element Mi;j represents the probability to transit from the assume that the state transitions of adjacent wireless links
ith state to the jth state. So, M is a Cðn þ 2Þ  Cðn þ 2Þ are independent. The probability that R remains alive at
matrix. For example, the matrix M corresponding to Fig. 6 is time unit t can be written as
briefly shown in Fig. 7.
Matrix M represents the state transition probabilities P3 ðR; tÞ ¼ ð1  P1 ði1 ; tÞÞ  ð1  P1 ði2 ; tÞÞ     
after one time unit. It is a simple result that the kth power of ð1  P1 ðik ; tÞÞ
M, denoted as M k , represents the state transition probabil- Y
k
k
ities after k time units. That is, Mi;j is the probability that a ¼ ð1  P1 ðij ; tÞÞ:
link at state i transits to state j after k time units. j¼1
Next, we will develop several probabilistic functions.
It follows that the probability that R has become broken at
Suppose that a wireless link is in state i at time unit 0. Let’s
denote by P1 ði; tÞ the probability that the link has become time t is
broken at time unit t. By matrix M, we can easily derive that P4 ðR; tÞ ¼ 1  P3 ðR; tÞ:
X
t
P1 ði; tÞ ¼ Mi;j : Let P5 ðR; tÞ be the probability that R remains alive at time
j2 layer nþ1;nþ2 unit t  1, but becomes broken at time unit t. We have
Furthermore, we need to know the probability that a 8
< P4 ðR; tÞ if t ¼ 1
wireless link is in state i at time unit 0, remains alive at time
P5 ðR; tÞ ¼ P4 ðR; tÞ  P4 ðR; t  1Þ if t > 1
unit t  1, and becomes broken at time unit t. Let’s denote :
¼ P3 ðR; t  1Þ  P3 ðR; tÞ:
by P2 ði; tÞ this probability. We can derive that
370 IEEE TRANSACTIONS ON MOBILE COMPUTING, VOL. 2, NO. 4, OCTOBER-DECEMBER 2003

Fig. 6. State transition diagram of a wireless link when n ¼ 5.

pffiffiffi
Finally, we conclude that the expected lifetime of route R, to be 10 meters (i.e., the radius of each cell is r ¼ 10= 3).
denoted as EðRÞ, is The number of layers (i.e., n) a radio can cover is 15 or 25.
Each time unit is 10 seconds. So, this is close to pedestrians’
X
1
EðRÞ ¼ P5 ðR; tÞ  t  ðlength of a time unitÞ: ð1Þ roaming speed. At the end of the section, we also present
t¼1 some simulation results by applying our lifetime prediction
scheme to DSR to observe how our prediction can help
4 NUMERICAL AND SIMULATION RESULTS improve routing efficiency (such as network throughput,
This section presents some numerical and simulation packet drop ratio, and routing control overhead).
results. Most current wireless LAN cards’ transmission 4.1 Determining the Level of Accuracy
distances range between one to a few hundred meters. So, Recall the route lifetime in (1). The summation has to be
we set the distance between the centers of two adjacent cells taken from t ¼ 1 to the infinity. This is computationally
infeasible due to the infinite upper bound. An upper bound
has to be set for t to calculate EðRÞ. In this simulation, we
randomly generate five routing paths with lengths 1, 3, 6, 9,
and 12 and calculate their expected lifetimes by setting t’s
upper bound from 100 to 1,000 (called tmax below). That is,
we approximate (1) by the following formula:

X
tmax
P5 ðR; tÞ  t  ðlength of a time unitÞ:
t¼1

Through such experiments, we hope to determine a

reasonable value for tmax .
The result is in Figs. 8a and 9a for n ¼ 15 and 25,
respectively. As can be seen, the expected route lifetime
converges when tmax enlarges. Longer routes’ lifetimes will
stabilize at smaller tmax than shorter routes’. When n ¼ 15,
most route lifetimes stabilize at tmax ¼ 500, while when
n ¼ 25, most route lifetimes stabilize at tmax ¼ 1; 000. So, we
will set tmax ¼ 500 and 1,000 when n ¼ 15 and 25,
Fig. 7. The state transition matrix M when n ¼ 5. respectively, for the rest of the simulations.
TSENG ET AL.: ON ROUTE LIFETIME IN MULTIHOP MOBILE AD HOC NETWORKS 371

Fig. 8. The expected route lifetime versus tmax when n ¼ 15.

Fig. 9. The expected route lifetime versus tmax when n ¼ 25.

Figs. 8b and 9b show the computational costs in the routing paths of lengths 3, 6, 9, and 12 were generated on a
above simulations. The costs all increase linearly with cellular plane. Each host on the routes roamed around
respect to tmax . Depending on the value of tmax , around a randomly following the same discrete-time, random-walk
few minutes are required for the computation.1 Our model. After each time unit, we then checked whether the
analysis shows that the main cost is on calculating the corresponding route still remained alive or not. This was
powers of matrix M. The size of M is Cðn þ 2Þ  Cðn þ 2Þ repeated until the route was broken, and then we recorded
(for example, when n ¼ 15, Cðn þ 2Þ ¼ 4; 913). By a typical the lifetime of the route. For each route, 20,000 such
row-by-column matrix multiplication, each multiplication simulation runs were executed and we took the average
has time complexity OðCðn þ 2Þ3 Þ. Since each M t has to be of these route lifetimes.
calculated, t ¼ 1 . . . tmax , the overall time complexity is Figs. 10 and 11 compare the probability distribution of
Oðtmax  Cðn þ 2Þ3 Þ. route breakage time (i.e., P5 ðR; tÞ) obtained from such
random simulation against our numerical analysis for
4.2 Verifying Numerical Results by Simulations several different routing paths when n ¼ 15 and 25,
To verify the correctness of our analytic results, we have respectively. The route being simulated is listed on the
also conducted some randomized simulations. Random top of each illustration. As we can see, the analytic results fit
pretty well with the simulation results. From these figures,
1. The simulations were run in an IBM-compatible PC with AMD
XP1600+with 256 MB DRAM memory. The Operating System was we can also see how route lifetime degrades as routes
Windows 2000 pro. become longer.
372 IEEE TRANSACTIONS ON MOBILE COMPUTING, VOL. 2, NO. 4, OCTOBER-DECEMBER 2003

Fig. 10. Comparison of route breakage probability distribution, P5 ðR; tÞ, obtained from random simulation and analysis when n ¼ 15 with route length
equal to: (a) 3 links, (b) 6 links, (c) 9 links, and (d) 12 links.

4.3 Application 1: Determining the Proper Size lifetime. The results are shown in Figs. 12 and 13. We see
of a MANET that route lifetimes all degrade significantly from one to five
In the literature, a lot of efforts have been devoted to hops. So, practical sizes of MANETs would range within
improve the scalability of routing protocols for MANETs. around five hops.
While this is definitely important, it remains a question
4.4 Application 2: Choosing Proper Routing Paths
whether it is cost-effective to have a very large MANET.
Intuitively, long routing paths with large hop counts may Given multiple paths between a pair of source and
suffer higher route breakage probability. So, more route- destination hosts, different criteria may be used to choose
searching efforts may be incurred if one intends to maintain a proper route. Hop counts are probably the most widely
longer routing paths. Our result provides a formal model, used criteria to choose routes. The proposed result can be
under the random walk assumption, to determine the used to evaluate routes based on their lifetime.
In this experiment, given a fixed pair of source and
relationship between route lifetime and route length.
destination, we uniformly place a number of hosts between
In this experiment, we generate multiple routes by
them which are spaced by the same distance. The route
varying their lengths and states. Specifically, we consider forms a straight line. Fig. 14 shows one experiment with the
routes of length i with state ½hj; 0i; hj; 0i; . . . ; hj; 0i
, where source at cell (0, 0) and destination at cell (50, 0) with
i ¼ 1::20 and j ¼ 3, 6, 9, and 12 (for n ¼ 15) or j ¼ 5, 10, 15, n ¼ 15. When the route length is h, the ith host will be
and 20 (for n ¼ 25). Then, we evaluate the expected route placed at cell ðb50i
h c; 0Þ, i ¼ 1::h  1. We then evaluate the
TSENG ET AL.: ON ROUTE LIFETIME IN MULTIHOP MOBILE AD HOC NETWORKS 373

Fig. 11. Comparison of route breakage probability distribution, P5 ðR; tÞ, obtained from random simulation and analysis when n ¼ 25 with route length
equal to: (a) 3 links, (b) 6 links, (c) 9 links, and (d) 12 links.

Fig. 12. The expected route lifetime versus route length when n ¼ 15. Each route consists of links of the same state.

expected route lifetime for different values of h. As can be After h  12, there is almost no gain in terms of route
seen, the expected lifetime increases rapidly when h ranges lifetime because routes have too many links.
between 4 to 10. At this stage, more hops are very helpful Fig. 15 shows a similar experiment with the source at cell
for route lifetime because hosts are closer to each other. (0, 0) and destination at cell (100, 0) with n ¼ 25. The trend
374 IEEE TRANSACTIONS ON MOBILE COMPUTING, VOL. 2, NO. 4, OCTOBER-DECEMBER 2003

Fig. 13. The expected route lifetime versus route length when n ¼ 25. Each route consists of links of the same state.

Fig. 14. The expected route lifetime versus route length with fixed source and destination hosts when n ¼ 15. The relay hosts are spaced regularly
and form a straight line.

Fig. 15. The expected route lifetime versus route length with fixed source and destination hosts when n ¼ 25. The relay hosts are spaced regularly
and form a straight line.

is similar: When h ranges from 12 to 18, the expected route 2 M bits/sec and a transmission distance of 250m. The cell
lifetime increases sharply, after which the route lifetime diameter is 10m. So, the central points of two neighboring
even degrades. pffiffiffi
cells are distanced by 5 3m, and each radio hop is around
4.5 Application 3: Improving the Routing Efficiency pffiffi cells. Each simulation lasts 1,000 sec. The numbers of
250
5 3
of DSR hosts simulated are: 30, 50, ..., 190. Each host is the source of
We also integrate our lifetime prediction algorithm into the one connection with a randomly selected destination. So, the
DSR routing protocol [8] to observe how our result can help number of connections being generated is equal to the
improve routing efficiency. The simulation platform is
number of hosts in the network. Each connection starts at a
Glomosim 2.03 [20]. The original DSR adopts shortest-path
routing. Our scheme will choose the path with the longest random time and sustains for an interval of 100 sec. For each
expected lifetime. We compare throughput, packet drop connection, packets are generated with a constant rate of
ratio, and routing overhead. 2 packets/sec. Each packet is of size 1,024 bytes. Intuitively,
The simulation environment is as follows: The network we do not change the offered traffic load of each connection,
area is 1,000m x 1,000m. Each radio has a transmission rate of but we tune the number of connections to vary the load.
TSENG ET AL.: ON ROUTE LIFETIME IN MULTIHOP MOBILE AD HOC NETWORKS 375

Fig. 16. Performance comparison: (a) network throughput, (b) packet drop ratio, and (c) routing overhead (DSR adopts shortest-path routing, while
“Lifetime” chooses routes with the longest expected lifetimes).

Hosts move following the random-way-point model by Fig. 16c compares the routing overheads. As expected, DSR
interleaving between roaming for 1 second and pausing for spends more control packets to recover and maintain
10 seconds. The roaming speed is 10 m/sec, so this is close to connections. With our lifetime prediction algorithm, there
a random walk, which changes cells at the rate of one cell per are less route recoveries and, thus, less control packets. The
11 seconds. In route search, a destination will wait for 5ms result verifies the effectiveness of our result in real
after the first ROUTE_REQ arrives before responding to the applications.
corresponding source. In predicting route lifetime, we use
tmax ¼ 300 in the simulation. 5 CONCLUSIONS
Fig. 16a compares throughputs of the two protocols
In this paper, we have formally derived the probability
under different network densities. With lifetime prediction,
distribution of the lifetime of a given routing path in a
the saturated network throughput increases about 20 per-
cent compared that of the shortest-path routing. This MANET based on the discrete-time, random-walk model.
indicates that choosing the shortest path is not always the Most existing works predict the lifetime of a wireless link
best choice. Instead, reliability is more important. To based on simpler models and take the minimum lifetime of
understand the reason, we calculate the packet drop ratio, the wireless links comprised of a routing path as the route
which is defined to be the probability of a packet being lifetime. The result proposed in this work can be used in
deleted in the middle way even if it is sent correctly from a measuring the scalability of a MANET and in evaluating
source. The main reason for packet dropping is the routing paths during route selection procedure. Human’s
existence of broken connections before sources are informed roaming pattern cannot be characterized by a single
(this is done by ROUTE_ERROR packets in DSR). The
mobility model (for example, pedestrians, vehicles, or
simulation result is shown in Fig. 16b. Choosing the shortest
people of different job nature all exhibit different roaming
paths is sometimes dangerous because they are likely to
become broken pretty soon. Not until a source is informed patterns). Route lifetime prediction is still a challenging
of a broken path, a lot of useless packets will be sent from question for roaming patterns other than the random walk
that source, wasting a lot of bandwidth. Packets would model. Understanding this factor will greatly facilitate
experience longer delays too due to route recovery. Finally, routing protocols in choosing proper routes.
376 IEEE TRANSACTIONS ON MOBILE COMPUTING, VOL. 2, NO. 4, OCTOBER-DECEMBER 2003

ACKNOWLEDGMENTS Yu-Chee Tseng received the BS and MS

degrees in computer science from the National
This work is cosponsored by the MOE Program for Taiwan University and the National Tsing-Hua
University in 1985 and 1987, respectively. He
Promoting Academic Excellence of Universities, Taiwan,
worked for D-LINK Inc. as an engineer in 1990.
under grant numbers A-91-H-FA07-1-4 and 89-E-FA04-1-4. He received the PhD degree in computer and
information science from the Ohio State Uni-
versity in January of 1994. From 1994 to 1996,
REFERENCES he was an associate professor in the Depart-
[1] S. Agarwal, A. Ahuja, J. Singh, and R. Shorey, “Route-Lifetime ment of Computer Science, Chung-Hua Univer-
Assessment Based Routing (RABR) Protocol for Mobile Ad-Hoc sity. He joined the Department of Computer Science and Information
Networks,” IEEE Int’l Conf. Comm., pp. 1697-1701, 2000. Engineering, National Central University in 1996 and was a full professor
[2] I.F. Akyildiz and J. Ho, “Dynamic Mobile User Location Update until 1999. Since August 2000, he has been a full professor in the
for Wireless PCS Networks,” ACM Wireless Networks, vol. 1, no. 2, Department of Computer Science and Information Engineering, National
pp. 187-196, July 1995. Chiao-Tung University, Taiwan. Dr. Tseng has served as a program
[3] I.F. Akyildiz, J. Ho, and Y.-B. Lin, “Movement-Based Location chair for the Wireless Networks and Mobile Computing Workshop, 2000
Update and Selective Paging for PCS Networks,” IEEE/ACM and 2001, as an associate editor for The Computer Journal, as a guest
Trans. Networking, vol. 4, no. 4, pp. 629-638, Aug. 1996. editor for ACM Wireless Networks special issue on “Advances in Mobile
[4] R. Dube, C. Rais, K. Wang, and S. Tripathi, “Signal Stability Based and Wireless Systems,” as a guest editor for IEEE Transactions on
Adaptive Routing (SSA) for Ad-Hoc Mobile Networks,” IEEE Computers special issue on “Wireless Internet,” as a guest editor for
Personal Comm., pp. 36-45, Feb. 1997. Journal of Internet Technology special issue on “Wireless Internet:
[5] Z. Haas and M. Pearlman, “ZRP: A Hybrid Framework for Applications and Systems,” as a guest editor for Wireless Communica-
Routing in Ad Hoc Networks,” Ad Hoc Networking, C.E. Perkins, tions and Mobile Computing special issue on “Research in Ad Hoc
ed., chapter 7, Addison-Wesley, 2000. Networking, Smart Sensing, and Pervasive Computing,” as an editor for
[6] J. Ho and I.F. Akyildiz, “Mobile User Location Update and Paging Journal of Information Science and Engineering, and as a guest editor
under Delay Constraints,” ACM Wireless Networks, vol. 1, no. 4, for Telecommunication Systems special issue on “Wireless Sensor
pp. 413-426, Dec. 1995. Networks.” He received the Outstanding Research Award, 2001-2002,
[7] M. Jiang, J. Li, and Y. Tay, “Cluster Based Routing Protocol from the National Science Council, ROC. His research interests include
(CBRP) Functional Specification,” internet draft, 1998. mobile computing, wireless communication, network security, and
[8] D.B. Johnson, D. Maltz, and J. Broch, “DSR: The Dynamic Source parallel and distributed computing. Dr. Tseng is a senior member of
Routing Protocol for Multihop Wireless Ad Hoc Networks,” Ad the IEEE.
Hoc Networking, C.E. Perkins, ed., chapter 5, Addison-Wesley,
2000. Yueh-Feng Li received the BS and MS
[9] S.-J. Lee, W. Su, and M. Gerla, “Ad Hoc Wireless Multicast with degrees in computer science engineering from
Mobility Prediction,” Proc. Int’l Conf. Computer Comm. and the National Central University, Chung-Li,
Networks, pp. 4-9, 1999. Taiwan, ROC, in 2000 and 2002, respectively.
[10] X. Lin and I. Stojmenovic, “GEDIR: Loop-Free Location Based Currently, he is an associate researcher in the
Routing in Wireless Networks,” Proc. IASTED Int’l Conf. Parallel Wireless Communication Technology Lab.,
and Distributed Computing and Systems, pp. 1025-1028, 1999. Telecommunication Laboratories, Chunghwa
[11] M. Mauve, J. Widmer, and H. Hartenstein, “A Survey on Position- Telecom Co., Ltd.
Based Routing in Mobile Ad Hoc Networks,” IEEE Network,
pp. 30-39, Nov./Dec. 2001.
[12] A.B. McDonald and T.F. Znati, “A Mobility-Based Framework for
Adaptive Clustering in Wireless Ad Hoc Networks,” IEEE
J. Selected Areas in Comm., vol. 17, no. 8, pp. 1466-1487, Aug. 1999. Yu-Chia Chang received the BS degree from
[13] A.B. McDonald and T.F. Znati, “Predicting Node Proximity in Ad- National Central University, Taiwan, in 2000.
Hoc Networks: A Least Overhead Adaptive Model for Selecting He is presently acquiring both the MS and
Stable Routes,” Proc. ACM Int’l Symp. Mobile Ad Hoc Networking PhD degrees in the Department of Computer
and Computing, pp. 29-33, 2000. Science and Information Engineering, National
[14] K. Paul, S. Bandyopadhyay, A. Mukherjee, and D. Saha, Central University. He once participated in the
“Communication-Aware Mobile Hosts in Ad-Hoc Wireless Net- Program for Promoting Academic Excellence
work,” Proc. Int’l Conf. Personal Wireless Comm., pp. 83-87, 1999. of Universities, carried out by the Ministry of
[15] C. Perkins and P. Bhagwat, “Highly Dynamic Destination- Education, in 2000-2001. He has been an
Sequenced Distance-Vector (DSDV) Routing for Mobile Compu- associate in the Design and Implementation of
ters,” Proc. ACM SIGCOMM Symp. Comm., Architectures and Location-Aware Mobile Ad Hoc Networks, promoted by the National
Protocols, pp. 234-244, Sept. 1994. Science Council, the Executive Yuan of the ROC, since 2000. His
[16] E.M. Royer and C.-K. Toh, “A Review of Current Routing research interests include wireless ad hoc networks, wireless sensor
Protocols for Ad Hoc Mobile Wireless Networks,” IEEE Personal networks, and bluetooth technology.
Comm., pp. 46-55, 1999.
[17] W. Su, S.-J. Lee, and M. Gerla, “Mobility Prediction in Wireless
Networks,” Proc. IEEE Military Comm. Conf., vol. 1, pp. 491-495,
2000.
[18] C.K. Toh, “Associativity-Based Routing for Ad Hoc Mobile . For more information on this or any computing topic, please visit
Networks,” Wireless Personal Comm. J., vol. 4, no. 2, pp. 103-139, our Digital Library at http://computer.org/publications/dlib.
Mar. 1997.
[19] Y.-C. Tseng and W.-N. Hung, “An Improved Cell Type
Classification for Random Walk Modeling in Cellular Networks,”
IEEE Comm. Letters, vol. 5, no. 8, pp. 337-339, Aug. 2001.
[20] X. Zeng, R. Bagrodia, and M. Gerla, “GloMoSim: A Library for
Parallel Simulation of Large-Scale Wireless Networks,” Proc. 12th
Workshop Parallel and Distributed Simulations, 1998.