Content uploaded by Somprakash Bandyopadhyay
Author content
All content in this area was uploaded by Somprakash Bandyopadhyay
Content may be subject to copyright.
Improving End-to-End Delay through Load Balancing
with Multipath Routing in Ad Hoc Wireless Networks
using Directional Antenna
Siuli Roy1, Dola Saha1, Somprakash Bandyopadhyay1,
Tetsuro Ueda2, Shinsuke Tanaka2
1 Indian Institute of Management Calcutta, Joka, Calcutta 700104 INDIA
{siuli, dola, somprakash}@iimcal.ac.in
2 ATR Adaptive Communications Research Laboratories, Kyoto 619-0288 JAPAN
{teueda, shinsuke}@atr.co.jp
Abstract. Multipath routing protocols are distinguished from single-path proto-
col by the fact that they use several paths to distribute traffic from a source to a
destination instead of a single path. Multipath routing may improve system per-
formance through load balancing and reduced end-to-end delay. However, two
major issues that dictate the performance of multipath routing - how many paths
are needed and how to select these paths. In this paper, we have addressed these
two issues in the context of ad hoc wireless networks and shown that the success
of multipath routing depends on the effects of route coupling during path selec-
tion. Route coupling, in wireless medium, occurs when two routes are located
physically close enough to interfere with each other during data communication.
Here, we have used a notion of zone-disjoint routes to minimize the effect of
interference among routes in wireless medium. Moreover, the use of directional
antenna in this context helps to decouple interfering routes easily compared to
omni-directional antenna.
1 Introduction
Multipath routing protocols are distinguished from single-path routing by the fact that
they look for and use several paths to distribute traffic from a source to a destination
instead of routing all the traffic along a single path. Utilization of multiple paths to
improve network performance, as compared to a single path communication, has been
explored in the past [1,2]. Classical multipath routing has focused on the use of multi-
ple paths primarily for load balancing and fault tolerance. Load balancing overcomes
the problem of capacity constraints of a single path by sending data traffic on multiple
paths and reducing congestion by routing traffic through less congested paths. The
application of multipath techniques in mobile ad hoc networks seems natural, as it
may help to diminish the effect of unreliable wireless links, reduce end-to-end delay
and perform load-balancing [2]. In addition, due to the power and bandwidth limita-
tions, a routing protocol in ad hoc networks should fairly distribute the routing traffic
among the mobile hosts. However, most of the current routing protocols in this context
are single-path protocols and have not considered the load-balancing issue. An unbal-
anced assignment of data traffic will not only lead to congestion and higher end-to-end
delay but also lead to power depletion in heavily loaded hosts. An on-demand mul-
tipath routing scheme is presented in [3], where alternate routes are maintained, so that
they can be utilized when the primary one fails. However, the performance improve-
ment of multipath routing on the network load balancing has not been studied exten-
sively. The Split Multipath Routing (SMR), proposed in [6], focuses on building and
maintaining maximally disjoint multiple paths.
Two key issues that dictate the performance of multipath routing are - how many
paths are needed and how to select these paths. In this paper, we have addressed
these two issues in the context of ad hoc wireless networks. It is shown that the per-
formance of multipath routing through proper load balancing improves substantially, if
we consider the effect of route coupling and use directional antenna instead of omni-
directional antenna with each user terminal. In the context of ad hoc networks, the
success of multipath routing depends on considering the effects of route coupling
during path selection. In [5], the effect of route coupling on Alternate Path Routing
(APR) in mobile ad hoc networks has been explored. It was argued that the network
topology and channel characteristics (e.g., route coupling) can severely limit the gain
offered by APR strategies. Route coupling is a phenomenon of wireless medium
which occurs when two routes are located physically close enough to interfere with
each other during data communication. As a result, the nodes in multiple routes con-
stantly contend for access to the medium they share and can end up performing worse
than a single path protocol. Thus, node-disjoint routes are not at all a sufficient condi-
tion for improved performance in this context.
In this paper, we use a notion of zone-disjoint routes in wireless medium where
paths are said to be zone-disjoint when data communication over one path will not
interfere with data communication along other paths. Our basic multipath selection
criterion for load balancing depends on zone-disjointness. However, getting zone-
disjoint or even partially zone-disjoint routes in ad hoc network with omni-directional
antenna is difficult, since the communication zone formed by each transmitting node
with omni-directional antenna covers all directions. Hence, one way to reduce this
transmission zone of a node is to use directional antenna. It has been shown that the
use of directional antenna can largely reduce radio interference, thereby improving the
utilization of wireless medium and consequently the network throughput [7]. In our
earlier work, we have developed the MAC and routing protocol using directional
ESPAR antenna [7] and demonstrated the performance improvement. In this paper, we
have investigated the effect of directional antenna on path selection criteria for mul-
tipath routing and obtained a substantial gain in routing performance through load
balancing using multiple paths with directional antenna.
The paper is organized as follows. In section 2, we define the notion of zone dis-
jointness and propose multipath selection criteria based on this notion. In section 3, we
evaluate the performance of proposed mechanism in a simulated environment to show
the effectiveness of our algorithm using directional antenna, followed by concluding
remarks in section 4.
2 Selection of Paths for Multipath Routing
2.1 Zone Disjoint Routes with Omni-directional and Directional Antenna
Most of the earlier works on multipath routing in ad hoc networks try to find out mul-
tiple node-disjoint/ maximally node-disjoint paths between source s and destination d
for effective routing with proper load balancing. [5,6,7]. Two (or multiple) paths be-
tween s and d are said to be node-disjoint, when they share no common node except s
and d. However, because of route coupling in wireless environment, node-disjoint
routes are not at all a sufficient condition for improved performance in this context.
Suppose, two sources, s1 and s2 are trying to communicate data to destinations, d1 and
d2 respectively. Let us assume that two node-disjoint paths are selected for communi-
cation- s1-x1-y1-d1 and s2-x2-y2-d2. Since the paths are node-disjoint, the end-to-end
delay in each case should be independent of the other. However, if x1 and x2 and/or y1
and y2 are neighbors of each other, then two communications can not happen simulta-
neously (because RTS / CTS exchange during data communication will allow either x1
or x2 to transmit data packet at a time, and so on). So, the end-to-end delay between
any source and destination does not depend only on the congestion characteristics of
the nodes in that path. Pattern of communication in the neighborhood region will also
contribute to this delay. This phenomenon of interference between communicating
routes is known as route coupling. As a result, the nodes in multiple routes will con-
stantly contend for access to the medium they share and can end up performing worse
than a single path protocol.
In this paper, we use a notion of zone-disjoint routes in wireless medium where
paths are said to be zone-disjoint when data communication over one path will not
interfere with data communication along other paths. In other words, two (or multiple)
paths between s and d are said to be zone-disjoint, when route-coupling between them
is zero.
The effect of route coupling has been measured in [8] using a correlation factor h.
In this paper, the correlation factor of a node n in a path P, hn (P), is defined as the
number of active neighbors of n not belonging to path P, where active neighbors of n
is defined as those nodes within the transmission zone of n that are actively partici-
pating in any communication process at that instant of time. For example, in figure 1,
S and D are communicating using two paths: S-a-b-c-D and S-d-e-f-D. So, all are
active nodes in this context of communication. Now, the active neighbors of node a is
{S, d,e,b}. So, correlation factor of node a in path {p= S-a-b-c-D}, ha (p)= number of
active neighbors not belonging to path p, i.e. 2.
The correlation factor h of path P, h (P), is defined as the sum of the correlation
factor of all the nodes in path P. When h (P) =0, path P is said to be zone-disjoint with
all other active paths, where active paths are those paths participating in
Fig. 1. . Two node-disjoint path with h = 9
communication processes at that instant of time. Otherwise, the path P is h–related
with other active paths.
Route coupling has a serious impact on path selection for load balancing via multi-
ple path. Let us refer figure 1 and assume that source S is sending data traffic to desti-
nation D along the path {S,a,b,c,D}. If S selects another path {S,d,e,f,D}, which is
closely coupled with the first path (as shown), and tries to distribute traffic across both
the path for load balancing, it may not result in performance improvement. In fact, it
has been observed that larger the correlation factor, the larger will be the average end-
to-end delay for both paths [8]. This is because two paths with larger correlation factor
have more chances to interfere with each other’s transmission due to the broadcast
feature of radio propagation. In addition, larger the correlation factor, the larger will
be the difference of end-to-end delay along multiple paths [8]. Based on this study, it
can be concluded that the path selection criterion for multipath routing in ad hoc net-
work needs to consider the correlation factor among multiple routes. In an environ-
ment of multiple communication among several source-destination pairs, even if a
path is less-loaded, that path may not be a good candidate for distributing traffic, if the
route coupling of that path with respect to other active paths is high. One way to alle-
viate this problem is to use zone-disjoint routes or maximally zone disjoint route for
load balancing. However, it is difficult to get fully zone-disjoint routes using omni-
directional antenna. As in figure 1, since both a and d are within omni-directional
transmission range of S, a RTS from S to node a will also disable node d. Similarly,
since both c and f are within omni-directional transmission range of D, a CTS from D
will disable both c and f. So, even if {a,b,c} and {d.e.f} are zone-disjoint, the lowest
possible h in case of omni-directional antenna with two multipath between s and d is
2. We call it minimal correlation factor hmin. With directional antenna, it is possible to
de-couple these two routes, making them fully zone-disjoint. For example, if each of
the nodes in figure 1 uses directional antenna where each node sets their transmission
beam towards its target node only, then the communication between S-a-b-c-D will
not affect the communication between S-d-e-f-D. Hence hmin(omni)=2 whereas
hmin(dir)=0. This will be further illustrated in the next section.
f
e
d
c
b
a
D
S
2.2 Number of Paths in a Multipath Route
Even if we get multiple zone-disjoint routes with minimal correlation factor
[hmin(omni)=2] using omni-directional antenna, the best-case packet arrival rate at the
destination node will be 1 packet at every 2*tp, where tp is the average delay per hop
per packet of a traffic stream on the path p. The best-case assumption is, traffic stream
in the network from S to D only with error-free transmission of packets. In contrast, if
we use directional antenna, best-case packet arrival rate at destination will be one
packet at every tp. It was illustrated analytically in [10] that the destination D will re-
ceive packets in alternate time-tick with omni-directional antenna and even if we in-
crease the number of paths between s and d beyond 2, the situation will not improve.
However, with directional antenna, when node a is transmitting a packet to node b,
S can transmit a packet to node d simultaneously. Thus, destination D will receive a
packet at every time-tick with two zone-disjoint paths using directional antenna. It is
to be noted here that two zone-disjoint paths with directional antenna are sufficient to
achieve this best-case scenario [10].
2.3 Selection of Paths Based on Correlation Factor h
Till now, we have considered communication over single s-d pair. However, situation
will deteriorate, if we consider multiple s-d pairs, engaged in simultaneous communi-
cations. Let us assume that each s-d pair selects two paths between them with lowest
possible h between them for effective load balancing. However, in the context of
multiple s-d pairs, even if two multipaths between, say, s1 and d1 are zone disjoint,
they may be coupled with other active routes between, say, s2 and d2. So, it is impera-
tive to consider all active routes and to find out h for each of them with respect to
other active routes in order to determine maximally zone-disjoint multipath between a
s-d pair such that it is not only maximally zone-disjoint with respect to each other but
also with respect to all active routes in the system.
However, it is a difficult task in the dynamic environment of ad hoc networks with
changing topology and communication pattern. An approximate solution to alleviate
this difficulty will be discussed in the next section. In this section, we will discuss the
mechanism of finding maximally zone disjoint multipaths with multiple s-d pairs and,
in the next section, will show the effectiveness of directional antenna over omni-
directional antenna in this context. To do this, initially we have assumed a static sce-
nario in our simulation environment. It has been assumed that each node is aware of
the topology and the communication pattern in the network. We use the following
algorithm to find out maximally zone-disjoint path between s-d:
Step I: Find out all node-disjoint paths between a s-d pair with number of hops H less
than Hmax (=5 in this experiment).
Step II: Find out h for each path between that s-d pair with respect to other active
paths.
Step III: Discard the path with highest h
Step IV: Repeat the process from Step II to step III with remaining paths between that
s-d pair until number of paths between them is two. These two paths are maximally
zone-disjoint path between that s-d pair.
2.4 Additional Criterion For Path Selection: Hop Count
However, zone disjointness alone is not sufficient for performance improvement. Path
length is also another important factor in multipath routing. A longer path with more
number of hops (H) will increase the end-to-end delay and waste more bandwidth. So,
even if a longer bypass route between a s-d pair has a low h, it may not be very effec-
tive in reducing end-to-end delay. To deal with this problem, our route-selection crite-
ria would be to minimize the product of h and H. Minimizing this factor will result in
maximally zone-disjoint shortest path. We call this factor g (=h*H). We use the fol-
lowing algorithm to find out maximally zone-disjoint shortest path between s-d:
Step I: Find out all node-disjoint paths between a s-d pair with number of hops H less
than Hmax (=5 in this experiment).
Step II: Find out h for each path between that s-d pair with respect to other active
paths.
Step III: Find out g for each path between that s-d pair.
Step IV: Discard the path with highest g
Step V: Repeat the process from Step II to step IV with remaining paths between that
s-d pair until number of paths between them is two. These two paths are maximally
zone-disjoint shortest path between that s-d pair.
3 Multipath Routing Performance
The proposed mechanism has been evaluated on a simulated environment under a
variety of conditions to estimate the basic performance. In the simulation, the envi-
ronment is assumed to be a closed area of 1500 x 1000 square meters in which mobile
nodes are distributed randomly. We present simulation results for networks with 40
mobile hosts, operating at a transmission range of 350 m. In order to evaluate the
effect of changing topology due to mobility, several snap-shots of random topology
with varying source-destination pairs are considered during our experiments. The
effective width of directional beam in case of directional antenna is assumed to be 60º.
In order to implement any routing protocol using directional antenna, each node
should know the best possible directions to communicate with its neighbors. So, each
node periodically collects its neighborhood information and forms a Neighborhood-
Link-State Table (NLST) at each node [7]. Through periodic exchange of this NLST
with its neighbors each node becomes aware of the approximate global network topol-
ogy and NLST at each node is upgraded to GLST (Global Link State Table). A direc-
tional MAC protocol, as discussed in our earlier work [7], has been implemented in
our simulator using information kept in GLST. Implementation of omni-directional
MAC follows the basic IEEE 802.11 scheme. A modified link-state routing protocol,
based on our earlier work [7,9] has been implemented. In the context of directional
antenna, GLST not only depicts the connectivity between any two nodes but also
maintains the best possible directions to communicate with each other. Moreover,
each node periodically propagates its knowledge about active-node-list, a list contain-
ing the node-ids of all nodes involved in any communication process at that instant of
time. It is to be noted that the perception of each node about the network topology or
number of active nodes in the network is only approximate. However, periodic re-
computation of routes by each intermediate node on a path will adaptively adjust itself
to the changing scenario.
Whenever a source s wants to communicate with a destination d, it computes mul-
tiple node-disjoint routes from s to d. From these multiple routes, it consults active
node list and computes maximally zone-disjoint multipath, or, maximally zone-disjoint
shortest multipath between s-d (as illustrated in previous section). However, due to
mobility and slow information percolation, it may not be possible for a source node to
perfectly compute maximally zone-disjoint multipath between s-d. To improve per-
formance under this condition, each intermediate node periodically recomputes the
same and adaptively modifies its routing decision.
We have compared the performance of (i) unipath routing with shortest path using
omni-directional antenna, (ii) maximally zone disjoint multipath with directional
and omni-directional antenna, and (iii) maximally zone disjoint shortest multipath
with directional and omni-directional antenna. In order to evaluate the impact of of-
fered load, we have experimented with 5, 10, 15 and 20 simultaneous communication.
Observations are recorded for 20 snap-shots in each case and the average values of
different parameters are computed. We evaluate the performance according to the
following metrics:
Load balancing: To measure load balancing in each case, we observe the number of
data packets forwarded at each node n [8]. If f(n) represents the number of data pack-
ets forwarded at each node n, the load balancing factor is the ratio of standard devia-
tion of f / mean of f, taken over all 40 nodes. Smaller the load balancing factor, better
the load balancing [8].
Average end-to-end delay: Average end-to-end delay per packet between a set of se-
lected s-d pairs is observed with increasing number of simultaneous communications
and with omni-directional and directional antenna. The timing assumptions are the
same as indicated in section 2.2.
Initially, to analyze average route-coupling among active routes, the experiment
starts with finding maximally zone-disjoint paths between selected s-d pairs. In order
to observe the impact of multiple simultaneous communications on route coupling
factor h, number of simultaneous communications are taken as 5, 10, 15 and 20. In
each case, we have found out average h, using omni-directional and directional an-
tenna. The result (figure 2) shows that the increase of h is much sharper in case of
omni-directional antenna. This implies that, as the number of s-d pair increases in the
system, the route-coupling among active routes increase much more sharply in
Fig. 2. Increase in route coupling with multiple multipath communications with omni-
directional and directional antenna
Fig. 3. Variation of load balancing factors with increasing number of simultaneous communi-
cations
case of omni-directional antenna compared to that with directional antenna. This has
an impact on end-to-end delay, as will be illustrated later.
As illustrated in figure 3, load balancing improves with increasing load with maxi-
mally zone disjoint multipaths, as compared to that with single shortest path. This
improvement is more pronounced when we use directional antenna. It is to be noted
that, smaller the load balancing factor, better the load balancing.
However, better load balancing does not imply better performance in this context.
Because of the possibility of high route coupling with omni-directional antenna (as
shown in figure 2), especially with increased number of simultaneous communication,
average end-to-end delay using multipath with omni-directional antenna will
Fig. 4. Variation of Average End-to-End Delay per packet with increasing number of simulta-
neous communications
[Multi-omni-eta: Multipath communication with maximally zone-disjoint path with omni-directional
Antenna
Multi-omni-gamma: Multipath communication with maximally zone-disjoint shortest path with omni-
directional Antenna
Multi-dir-eta: Multipath communication with maximally zone-disjoint path with directional Antenna
Multi-dir-gamma: Multipath communication with maximally zone-disjoint shortest path with directional
Antenna]
not show any significant improvement as compared to that with single path. Since
route coupling is far less with directional antenna, average end-to-end delay will be
substantially less with directional antenna than that with omni-directional antenna.
This is shown in figure 4. At the same time, path length is also another important
factor in multipath routing. A longer path with more number of hops (H) will increase
the end-to-end delay and waste more bandwidth. So, even if a longer bypass route
between a s-d pair has a low h, it may not be very effective in reducing end-to-end
delay. That is why, maximally zone disjoint shortest path with directional antenna will
show best performance, so far as both end-to-end delay and load balancing are con-
cerned (figure 4).
4 Conclusion
Multipath routing strategies in the context of ad hoc networks may improve load bal-
ancing, but may not improve system performance to the expected level through re-
duced end-to-end delay, unless we consider the effects of route coupling. However,
high degree of route coupling among multiple routes between any source and destina-
tion pair is inevitable, if we use omni-directional antenna. The situation will worsen, if
we consider multiple simultaneous communications with multipath routing scheme.
This paper has analyzed the problem and proposed a mechanism to alleviate the prob-
lem of route coupling using directional antenna. The paper also considers the advan-
tage of selecting maximally zone-disjoint as well as shorter route instead of longer by-
pass routes for effective load balancing and better network performance. Thus, all
active paths are maximally zone disjoint shortest paths. The final result shows that the
routing performance using multiple paths improves substantially with directional an-
tenna compared to that with omni-directional antenna.
Acknowledgement
This research was supported in part by the Telecommunications Advancement Or-
ganization of Japan.
References
1. Sajal K. Das, A. Mukherjee, Somprakash Bandyopadhyay, Krishna Paul, D. Saha, “Im-
proving Quality-of-Service in Ad hoc Wireless Networks with Adaptive Multi-path Rout-
ing, Proc. Of the GLOBECOM 2000, San Francisco, California, Nov. 27-Dec 1, 2000.
2. Aristotelis Tsirigos Zygmunt J. Haas, Siamak S. Tabrizi , Multi-path Routing in mobile ad
hoc networks or how to route in the presence of frequent topology changes , MILCOM
2001.
3. Nasipuri and S.R. Das, "On-Demand Multi-path Routing for Mobile Ad Hoc Networks,"
Proceedings of IEEE ICCCN'99, Boston, MA, Oct. 1999.
4. Johnson and D. A. Maltz, "Dynamic Source Routing in Ad Hoc Wireless Networks," T.
Imielinski and H. Korth, editors, Mobile Computing, Kluwer, 1996.
5. M. R. Pearlman, Z. J. Haas, P. Sholander, and S. S. Tabrizi, On the Impact of Alternate
Path Routing for Load Balancing in Mobile Ad Hoc Networks, MobiHOC 2000, p. 150, 3-
10.
6. S.J. Lee and M. Gerla, Split Multi-path Routing with Maximally Disjoint Paths in Ad Hoc
Networks, ICC 2001.
7. Siuli Roy, Dola Saha, S. Bandyopadhyay, T. Ueda, S. Tanaka. A Network-Aware MAC
and Routing Protocol for Effective Load Balancing in Ad Hoc Wireless Networks with Di-
rectional Antenna. ACM MobiHoc,2003, Maryland, USA, 1-3 June 2003.
8. Kui Wu and Janelle Harms, On-Demand Multipath Routing for Mobile Ad Hoc Networks
EPMCC 2001, Vienna, 20th – 22nd February 2001.
9. Romit RoyChouldhury, Somprakash Bandyopadhyay and Krishna Paul, “A Distributed
Mechanism for Topology Discovery in Ad hoc Wireless Networks using Mobile Agents”,
Proc. of the First Annual Workshop On Mobile Ad Hoc Networking & Computing
(MOBIHOC 2000), Boston, Massachusetts, USA, August 11, 2000.
10. Somprakash Bandyopadhyay, Siuli Roy, Tetsuro Ueda, Shinsuke Tanaka "Multipath
Routing in Ad Hoc Wireless Networks with Omni Directional and Directional Antenna: A
Comparative Study" Proc. of the IWDC, Calcutta, December 27-30,2003, (LNCS 2571),
Springer Verlag , Dec 2002.