We present a hybrid routing protocol for both pure and hybrid ad hoc networks which uses the mechanisms of swarm
intelligence to select next hops. Our protocol, Ad hoc Networking with Swarm Intelligence (ANSI), is a congestion-aware
routing protocol, which, owing to the self-organizing mechanisms of swarm intelligence, is able to collect more information
about the local network and make more effective routing decisions than traditional MANET protocols.
Driving Behavioral Change for Information Management through Data-Driven Gree...
A swarm intelligence-based unicast routing protocol for hybrid ad hoc networks
1. ARTICLE IN PRESS
Journal of Systems Architecture xxx (2006) xxx–xxx
www.elsevier.com/locate/sysarc
ANSI: A swarm intelligence-based unicast routing
protocol for hybrid ad hoc networks q,qq
Sundaram Rajagopalan *, Chien-Chung Shen
DEGAS Networking Group, Department of Computer and Information Sciences, University of Delaware,
Newark, DE 19716, USA
Abstract
We present a hybrid routing protocol for both pure and hybrid ad hoc networks which uses the mechanisms of swarm
intelligence to select next hops. Our protocol, Ad hoc Networking with Swarm Intelligence (ANSI), is a congestion-aware
routing protocol, which, owing to the self-organizing mechanisms of swarm intelligence, is able to collect more information
about the local network and make more effective routing decisions than traditional MANET protocols. Once routes are
found, ANSI maintains routes along a path from source to destination effectively by using swarm intelligence techniques,
and is able to gauge the slow deterioration of a link and restore a path along newer links as and when necessary. ANSI is
thus more responsive to topological fluctuations. ANSI is designed to work over hybrid ad hoc networks: ad hoc networks
which consist of both lower-capability, mobile wireless devices and higher-capability, wireless devices which may or may
not be mobile. In addition, ANSI works with multiple interfaces and with both wired and wireless interfaces.
Our simulation study compared ANSI with AODV on both hybrid and pure ad hoc network scenarios using both TCP
and UDP data flows. The results show that ANSI is able to achieve better results (in terms of packet delivery, number of
packets sent, end-to-end delay, and jitter) as compared to AODV in most simulation scenarios. In addition, ANSI achieves
this performance with fewer route errors as compared to AODV. Lastly, ANSI is able to perform more consistently, con-
sidering the lower variation (measured as the width of the confidence intervals) of the observed values in the results of the
experiments. We show that ANSI’s performance is aided by both its superior handling of routing information and also its
congestion awareness properties, though we see that congestion awareness in ANSI comes at a price.
Ó 2006 Elsevier B.V. All rights reserved.
Keywords: Swarm intelligence; MANET; Hybrid network; Hybrid routing; Congestion aware routing
1. Introduction
q
A section of this work was presented at ICAI 2005, June 2005, Hybrid ad hoc networks consist of a mixture of
Las Vegas, NV, USA. mobile, ad hoc network (MANET) nodes and nodes
qq
This work is supported in part by National Science Foun- which belong to highly capable infrastructure such
dation under grant ANI-0240398.
*
Corresponding author. Tel.: +1 302 831 1131; fax: +1 302 831
as mesh networks or cellular networks. The problem
8458. of hybrid ad hoc networks is to make these networks
E-mail address: rajagopa@cis.udel.edu (S. Rajagopalan). work efficiently without relying on pre-configured
1383-7621/$ - see front matter Ó 2006 Elsevier B.V. All rights reserved.
doi:10.1016/j.sysarc.2006.02.006
2. ARTICLE IN PRESS
2 S. Rajagopalan, C.-C. Shen / Journal of Systems Architecture xxx (2006) xxx–xxx
network topologies or centralized control. Hybrid of lower-level components. The combination/inter-
ad hoc networks are useful in many situations where action of lower-level components in SI such as
impromptu communication facilities are required positive/negative feedback and amplification of fluc-
such as battlefield communications, and disaster tuations along with multiple interactions are the
relief missions. mechanisms which allow a node to change routing
Since the problem of hybrid ad hoc networking information quickly and efficiently to adjust to an
shares a lot of problems with typical MANET prob- ever-changing local topology and route deteriora-
lems, typical routing solutions for hybrid networks tion, thus initiating fewer link breakages.
start with a MANET routing solution and then Our protocol, ANSI, uses a highly flexible cost
apply some optimizations to work for specific sce- function which allows it to use the information col-
narios. A number of ad hoc routing protocols have lected from the local ant activity, such as the conges-
been proposed, for example [1–5], of which some of tion status of the neighboring nodes, in useful ways.
them, like AODV [1] work on hybrid ad hoc net- In addition, the ant-like working of our protocol
works. In proactive protocols such as [5], nodes in allows for the maintenance of multiple routes to a
the network maintain routing information to all destination. In nodes which use proactive routing
other nodes in the network by periodically exchang- in ANSI, this fact is used to perform stochastic rout-
ing routing information. Nodes using reactive pro- ing, and in nodes that use perform reactive routing
tocols, such as [1,2], delay the route acquisition (pure MANET nodes), when one route fails, others
until a demand for a route is made. Hybrid proto- may be used. Our motivation comes from the fact
cols, like [4,6], use a combination of both proactive that different networks face different conditions,
and reactive activities to gather routes to the desti- and thus a protocol suite should allow for various
nations in a network—nodes using ZRP, for exam- configurations as the network conditions dictate.
ple, proactively collect routes in their zone, and Furthermore, supporting multiple routes simulta-
other routes are collected reactively. In [6], on the neously is essential to ensure survivability of the net-
other hand, the level of proactive activity and reac- work [9]. ANSI facilitates ad hoc unicast routing by
tive activity are chosen autonomously by the nodes exploiting route finding behaviors that are emergent
in the network, and proactive activity is only seen from ant packets working collectively, rather than
around favorite destination nodes. In most tradi- explicitly coding them to cope with the problem.
tional reactive protocols, like [1,2], only when a We formulate the routing problem at node i as a
route breaks irreparably does the protocol mecha- set of ‘‘food foraging’’ problems from nest i, where
nisms repair the damage. In reality, route deteriora- each ‘‘food source’’ is a destination d in the net-
tion in mobile networks is most often not sudden work. In this formulation, next hops are evaluated
but gradual,1 and most often available routes get on the basis of the strength of the pheromone trail2
better/deteriorate gradually and not suddenly. So on the link connecting a node and a next hop.
the routing protocol should continuously maintain The remainder of this paper is organized as fol-
information about the nodes in the local area to per- lows: In the next section, we discuss a number of
form effectively and avoid too may link breakages. approaches and protocols which are related to our
In this paper, we present a hybrid routing suite research. In Section 3, we describe in detail the com-
(with both proactive and reactive components) for ponents of ANSI unicast routing protocol, and fol-
hybrid ad hoc networks which uses the mechanisms low it with Section 4 where we discuss the results of
of swarm intelligence [7] to select good routes to des- the comparison of simulated models of ANSI with
tinations. We use Swarm Intelligence (SI) because SI a popular routing protocol, AODV [1]. We conclude
mechanisms allow for self-organizing systems [8] and in Section 5 with a brief note on future research effort.
maintain state information about the neighboring
network better than traditional MANET routing 2. Related work
mechanisms. Self-organizing systems are robust
environments where erroneous system behavior is The main ingredients of SI, positive/negative
corrected autonomously by the coordinated working reinforcement, and amplification of fluctuations
1
Some routes, such as routes to neighbors, break suddenly,
2
when the neighbors go out of range. We are commenting on the The computational equivalent of the chemical deposited on
general case here. the forest floor by ants.
3. ARTICLE IN PRESS
S. Rajagopalan, C.-C. Shen / Journal of Systems Architecture xxx (2006) xxx–xxx 3
are achieved in an environment with multiple inter- DBF—ants are not used as feedback agents to rein-
actions among nodes [7]. Because the above compo- force routes positively (in the case when a route is
nents of SI are the lower-level components required still good), negatively (when a route is no longer
for self-organizing behavior, the benefits of using good) or explore new routes randomly—ants in this
SI-based algorithms are not fully accrued if the approach are unicast to specified direction, not
any of the above lower-level components are not allowing for amplification of fluctuations, and
present in a swarm intelligence-based protocol. This depending on known metrics such as timestamp of
is because in any SI-based system, these aspects a route in the routing table.
work together in learning about the network. The approach used in [13] by Heissenbttel and
In [10] Baras and Mehta describe a swarm intel- Braun also relies on location information, and is a
ligence-based reactive ad hoc routing protocol purely proactive routing approach based on divid-
called PERA. PERA uses broadcast forward ants ing the network into logical zones and assigning
as exploratory agents sent out on-demand to find logical routers to each. Ants—forward ants and
new routes to destinations. Each ant holds a list of backward ants—are used by logical routers in this
nodes that were visited while exploring the network, approach to periodically check if the logical links
and since these ants are broadcast at each node, a connecting it to a randomly chosen destination are
forward ant can result in several backward ants— functional and reflect on the current state of the net-
ants sent by destination nodes in response to for- work surrounding the logical router. Positive and
ward ants. This uncovers several routes for each for- negative reinforcement are achieved by means of
ward ant sent, and at each node these multiple multiple interactions and pheromone additions (by
routes found to the destinations are maintained as forward and backward ants) and pheromone aging,
probability values. As with AntNet [11], the routing respectively. Random amplification of a new good
table Ri at node i is a probability matrix with a route in the face of topological fluctuations is
probability entry Pijd as the probability that a data possible by random dissemination of ants to desti-
packet at i’s FIFO queue will take the next hop j nations.
to be routed to d. Positive reinforcement is managed In [14], Gunes et al. outline ARA, a multipath,
in PERA using forward/backward ants and nega- purely reactive scheme. ARA uses forward ants
tive reinforcement is implicit—no explicit aging of and backward ants to create fresh routes from a
the pheromone trails is done. After a route has been node to a destination. When routes to a destination
established, PERA regularly uses forward ants to D are not known at S, a forward ant is broadcast,
find newer routes to destinations. This is wasteful, taking care to avoid loops and duplicate ants. When
considering the fact that forward ants cause a lot a forward ant is received at an intermediate node X
of network resources to be consumed and should via node Y, the ant reinforces the link XY in X to
not be sent when not necessary. route to all the nodes covered so far by the forward
In [12], Camara and Loureiro outline a source ant. When a forward ant is received at D, a back-
routing scheme in which the network relies on loca- ward ant is created which backtracks the path of
tion information and support from fixed infrastruc- the corresponding forward ant. At each node the
ture. Owing to a source routing approach, the backward ant is received, the link via which the
algorithm relies heavily on a source M destination backward ant is received is reinforced, like the for-
route which is available at the time of message cre- ward ant does, for all nodes which have been visited
ation. New nodes in the network start with using by the backward ant. In ARA, data packets per-
their neighbor’s routing table. The routing table, form the necessary (positive) reinforcement required
generated using shortest path algorithms, on the to maintain routes. When a path is not taken, it sub-
other hand, may contain information which is out- sequently evaporates (negative reinforcement) and
dated. Ants are unicast from a source to specific des- cannot be taken by subsequent data packets. Under
tinations, for example, the destination node may be the described scheme, amplification of topological
the node with the oldest information in the routing and network fluctuations is not possible except
table. This mechanism is used to make sure that the under extreme conditions when routes break
routing information in the source is updated and often.
recent. Thereby, ants are used in [12] with the In [15,16], Di Caro et al. describe AntHocNet, a
semantics of routing information updates, like clas- hybrid, stochastic approach to the routing problem
sical distance vector protocols such as DSDV or in MANET. AntHocNet is a congestion-aware pro-
4. ARTICLE IN PRESS
4 S. Rajagopalan, C.-C. Shen / Journal of Systems Architecture xxx (2006) xxx–xxx
tocol which only finds routes on-demand, but once a ing in immobile, highly capable infrastructure and
route is established, the route is proactively main- applies it only in those nodes, rather than letting
tained. This approach, argued by the authors to pure MANET nodes incur the costs due to the same
be more ant-like [16] than other competing ant- under high mobility conditions. Lastly, the flexible
based protocols, will fail to reduce overheads in very cost function (specifically, the congestion-awareness
high traffic/mobility scenarios, owing to the rate at property) in ANSI leverages the inherent nature of
which proactive ants are potentially unicast when swarm intelligence by collecting multiple routes
the mobility increases. This is because in high mobil- and using them to perform load balancing in all
ity/traffic scenarios, routes get invalidated often and sections of the network. This, as we will see, also
proactive activity has to increase appropriately to alleviates the tendency to create hotspots in the
keep a valid view of the network for routing, thus network.
increasing the load placed on the network. Indeed,
we do agree with the comment that the authors of 3. ANSI unicast routing protocol
AntHocNet make regarding the repeated path sam-
pling, and ANSI manages to steer clear from 3.1. Protocol overview
repeated path sampling by carefully choosing when
to engage in route discovery activity. ANSI is a hybrid routing protocol for hybrid ad
In [17], Wedde et al. present a new routing hoc networks comprising of both proactive and reac-
algorithm for energy efficient routing in mobile tive routing components. Pure MANET (mobile)
ad hoc networks. In their approach, they show nodes in ANSI use only reactive routing, and choose
that BeeAdHoc, a reactive source-routing protocol routes deterministically, while nodes belonging to
inspired by the foraging principles of honey bees, is more capable, infrastructured (immobile) networks
able to achieve energy consumption characteristics use a combination of both proactive and reactive
as compared to DSR, AODV and DSDV without routing and perform stochastic routing when multi-
compromising on traditional performance metrics ple paths are available. The outline of the process
such as packet delivery and throughput. of ANSI routing is as follows:
Our protocol, ANSI, is a hybrid protocol pro-
posed for hybrid ad hoc networks. Some character- 1. When a route to a destination D is required, but
istics seen in traditional on-demand routing not known at a node S, S broadcasts a forward
protocols can be seen in ANSI. For example, an reactive ant to discover a route to D.
optimization used in AODV, expanding ring search, 2. When D receives the forward reactive ant from S,
is also used in ANSI, albeit more efficiently, owing it source-routes a backward reactive ant to the
to the use of history information. Unlike traditional source S. The backward reactive ant updates
MANET protocols which engage in route mainte- the routing table of all the nodes in the path from
nance/discovery activity only when links break, S to D, allowing for data transfer from S to D.
ANSI continuously updates a node’s neighborhood 3. When a route fails at an intermediate node X, X
information using data packets and control packets first checks if there are other routes which can be
to alleviate the negative effects due to flooding the used to route the packet to D. If not, then ANSI
network with route discovery/maintenance. In addi- buffers the packets which could not be routed
tion, unlike traditional MANET protocols, ANSI and initiates a route discovery to find D by using
has a flexible cost function which allows it to per- a forward reactive ant to perform local route
form metric-centered routing. In our implementa- repair. Additionally, X sends a route error mes-
tion, we have performed congestion-aware routing, sage back to the source node S.
but it is easy to see how this cost function can be 4. Nodes belonging to more capable, infrastructured
modified to perform, say, energy efficient routing. networks maintain routes to their connected com-
When compared to other ant algorithms for ponents proactively, by periodic routing updates
MANET routing, we note that to the best of our using proactive ants. Nodes belonging to more
knowledge, there exists no other ant algorithm for capable, infrastructured networks also use sto-
hybrid ad hoc networks, but ANSI is able to per- chastic routing when multiple paths are available.
form well in both pure MANET and hybrid ad In addition, each node in the infrastructure
hoc networks. In addition, the ANSI design under- collects information about which mobile nodes
stands the advantages of proactive/stochastic rout- are connected to which infrastructure node.
5. ARTICLE IN PRESS
S. Rajagopalan, C.-C. Shen / Journal of Systems Architecture xxx (2006) xxx–xxx 5
5. When a route at D is known at a MANET node maintains a row for the destination-next hop
S, ANSI deterministically chooses the best next pair (d, j) along with the sijd (t), gijd, wijd, and aijd
hop to reach the destination. If S is part of a values described below:
highly capable infrastructure, then S may choose (a) sijd(t) is the pheromone trail concentration
to perform stochastic routing to the destination left on a trail ij used as a first hop to desti-
D, depending on the availability of multipath nation d at current time t due to all the ants
routes. that have traversed the trail, taking into
consideration the pheromone evaporation
We claim ANSI will perform better than typical (see Eq. (6)). s is thus a weighted measure
MANET protocols because of the working of the of how many times the trail ij was traversed
SI mechanisms at each node, which maintain rout- by packets intended to d and is thereby a
ing information and local information more effec- measure of the goodness of trail ij.
tively than traditional MANET routing protocols. (b) gijd is the heuristic value of going from j to i.
In addition, the congestion-awareness of ANSI also In our mapping, g is a measure of the dis-
helps in controlling the extent of congestion in high tance to the destination, distijd, going from
traffic scenarios. Lastly, in hybrid networks, ANSI i to d, when using next hop j. We set
1
is able to leverage the power of nodes belonging gijd ¼ 1 þ distijd .
to more capable networks to assist in routing activ- (c) wijd 2 [0, 1] is the value of the congestion
ities of the network. In Sections 3.3.1–3.3.4, we status at node j. If wijd = 1, then, node j is
explain the details of the above actions, and show considered not congested, and if wijd = 0,
how the SI mechanisms work at each node in main- then the node j is considered congested.
taining routing information. The value of w at a node j is measured as
the ratio of empty space in packets in the
3.2. Protocol model IP queue size to the number of packets
already in the IP queue at j.
3.2.1. Data structures (d) We see that the goodness of a next hop j is
Data structure (1) below is the ant structure, car- directly proportional to sijd(t), inversely
ried by all ants, and data structures (2) and (3) proportional to distijd and directly propor-
below are maintained at each node, and are updated tional to wijd. Thus, we write:
every time an ant arrives at the node. a
aijd ¼ ðcs Á sijd ðtÞ Þ Á ðcg Á gb Þ Á ðcw Á wc Þ
ijd ijd
(1) Ant structure: The following information is car- ð1Þ
ried by an ant p: where cs > 0, cg > 0, and cw > 0 are arbi-
(a) The ant ID of the ant, which is the (node trary constants, and a, b, c are integers such
ID, sequence number) pair. that a, b, c > 0.
(b) The number of nodes, m, which p visits, For our use, we need to normalize the
including the node p originated from. above value of aijd so that we may gauge
(c) The nodes-visited-stack (adapted from the relative effectiveness of each next hop.
[11]), Sp , containing information about We normalize it such that aijd 2 [0, 1]:
nodes V = {v1, v2, . . . , vm}, that can be a
reached by backtracking the ant p’s move- ðcs Á sijd ðtÞ Þ Á ðcg Á gb Þ Á ðcw Á wc Þ
ijd ijd
aijd ¼P a b c
ment (using the nodes-visited-stack), and l2J ðcs Á sild Þ Á ðcg Á gild Þ Á ðcw Á wild Þ
(d) The pheromone amount at v 2 V, pv. ð2Þ
(2) Ant decision table at node i, Ai : (adapted from
[18]). An ant decision table is a data structure where J is the set of next hops at i to desti-
that stores pheromone trail information for nation d. We then set cs = cg = cw = 1, and
routing from node i to a destination d via k pos- arrive at
sible next hop nodes J = {j1, j2, . . . , jk}. The link a b c
½sijd ðtÞŠ ½gijd Š ½wijd Š
ij in the ANSI network, between two nodes i aijd ðtÞ ¼ P a
ð3Þ
l2J ½sild ðtÞŠ ½gild Šb ½wild Šc
and j is assumed to be bidirectional. Routing
tables are computed from ant decision tables. where a, b and c are chosen appropriately
Each ant decision table entry Aijd for node i (see Section 4). The above formula was
6. ARTICLE IN PRESS
6 S. Rajagopalan, C.-C. Shen / Journal of Systems Architecture xxx (2006) xxx–xxx
adapted from Ant Colony Optimization ing the siJ 0 V 0 values on the trails iJ 0 that were
techniques outlined in [18]. The intuition negatively reinforced, i.e., no ants that traveled
behind this equation is that we want to V 0 ¼ fv01 ; v02 ; . . . ; v0m0 g were received) and recalcu-
use the metrics of hop distance and path lated, the aijv values for all entries V in Sp0 are
goodness and allow some flexibility as to recomputed and the new best next hops to destina-
how much we rely on either metric by vary- tions V are computed again. This is followed by an
ing a, b and c values. update of the routing table at node i. Negative rein-
As soon as an ant p is received at a node i via forcement of routes also happens when a route is
neighbor node j, i has the information about j’s con- explicitly invalidated by a route error message.
gestion status from Sp . The pheromone sp depos-
ijv (3) Routing table: The routing table at node i is a
ited by an ant p and the heuristic gijv to a table containing an entry for each destination
destination v in the ant p traversing from node j d reachable from node i along with the best next
to node i via nodes v 2 V are given by the equations: hop, Jd , to d. The best next hop, Jd , to a des-
i i
1 tination d is the next hop that contains the larg-
sp ¼
ijv ðv; i; j 2 V Þ ð4Þ est aijd value in Ai . The value of Jd is thereby
pj À pi i
updated every time an ant visits a node i. The
and routing table also contains the distance of d
1 from i in hops, and this information is used to
gp ¼ 1 þ
ijv ðv; i; j 2 V Þ ð5Þ set the number of hops for route discovery
depthðvÞ
when the routing table entry to d in i becomes
where pi and pj are the pheromone amounts of ant p defunct.In the case of nodes which are part of
at nodes i and j, respectively, and V = {v1, v2, . . . , highly capable infrastructure, the routing is sto-
vm} denotes the set of m nodes visited by p. The chastic, and the next hop is chosen directly from
value depth(v) is the depth of the node v in p’s nodes the ant decision table probabilistically. Specifi-
visited stack.All s values in Ai are evaporated cally, a next hop j at node i for destination d
according to Eq. (7) each time another ant, p 0 , visits is chosen with a probability of aijd.
node i. Let us say p 0 traverses the same trail ij at
time (t + D) as traversed by p at time t. p 0 then pos-
itively reinforces the trail ij "v 2 V in Sp . All other 3.2.2. Amplification of fluctuations
trails iJ 0 , (where J 0 is the set of all possible next hops The process of broadcasting ants during reactive/
from i except j) in the ant decision table Ai are not proactive route discovery/recovery/maintenance
positively reinforced, and in the event no ant finds new routes to nodes and alters the information
traverses through any of the other trails iJ 0 , the in the ant decision table accordingly. Because of the
trails iJ 0 eventually become invalidated (negatively nature of broadcast in the wireless medium, the
reinforced) owing to pheromone evaporation. The routes found as a result of forward reactive ant
new sijv at time (t + D) is calculated as follows: activity reflect the current status of the network
0 and accordingly amplify the current fluctuations in
sijv ðt þ DÞ ¼ evaporateðsijv ðtÞ; DÞ þ sp
ijv ð6Þ
the topology. Another mechanism amplifies the
p0 fluctuations in the local area: when a node receives
where sijv is the pheromone deposited on the trail
by p 0 over ij (see Eq. (4)). The function evapo- a unicast packet, it notes the neighbor node ID
rate(sijv(t), D) returns the pheromone amount left and reinforces the path to the source of the packet
on trail ij for destination v (after evaporation) due via the neighbor. In addition, when a data packet
to the ants which traversed ij before p 0 . The phero- is sent along a next hop, the node reinforces the next
mone evaporation model used to calculate how hop as a valid next hop to the destination. This
much of the earlier pheromone trail, sijv(t) is left be- mechanism also amplifies local fluctuations of net-
hind at (t + D) when p 0 traverses the trail ij is as work and topological characteristics and see to it
follows: that the nodes in the ANSI network use up-to-date
network and topological information.
sijv ðtÞ
evaporateðsijv ðtÞ; DÞ ¼ ð7Þ Some protocols, for e.g., [10], using SI mecha-
2D=c nisms for MANET argue for unicasting forward
where c is an arbitrary constant. After all the s val- reactive ants along one randomly chosen path to
ues in the ant decision table are evaporated (includ- the source and destination to amplify the fluctua-
7. ARTICLE IN PRESS
S. Rajagopalan, C.-C. Shen / Journal of Systems Architecture xxx (2006) xxx–xxx 7
tions in the network. Yet others, for e.g., [16], argue
for sending forward ants at regular intervals from
the source while the source is sending packets to
the destination. We feel that the above methods will
work in low traffic scenarios and in wired networks,
where there is little or no mobility, but not in highly
mobile MANET with high loads. Besides, we feel
that the right model for amplification of fluctuations
in a MANET using SI mechanisms is the model we
use: that of broadcasting forward ants only when a b
absolutely needed both at the source and intermedi-
ate nodes (to perform local repair), and using these Fig. 1. Local reinforcement in ANSI. (a) Reinforcement by data
packets. Node i, upon receiving a data packet from S via node j,
mechanisms with a neighbor discovery mechanism,
reinforces the path to node j via j and the source S via j.
and applying the rules of SI on the data collected (b) Reinforcement in neighbor discovery mechanisms. Upon
(viz., Eqs. (3)–(7)). By using the SI mechanisms receiving a HELLO beacon from j all nodes i reinforce trails via j.
appropriately in ANSI, we are able to reduce the
number of MAC layer resources used wastefully,
as well as be responsive to a network with high traf- hello packets are used to perform local route man-
fic rates, and provide better packet delivery rates agement by positively reinforcing previously known
and lower delay jitter characteristics. neighbors and new neighbors. The advantage of
using this mechanism can be explained as follows:
3.3. Protocol description If a direct route to a destination d is known at i
via this process, then a previously known indirect
A trail ij to destination d, sijd, is positively rein- route to d is less favored than the direct route by
forced in ANSI when (a) a new route to a destina- the reinforcement mechanisms in ANSI. Note that
tion d is found (via ant activity) at i via next hop HELLO messages are sent via all available interfaces
(neighbor node) j, and (b) when i uses an already to facilitate neighbor discovery over all possible
known nexthop node j again to route a packet to paths.
d. A trail ij is negatively reinforced when (a) the trail
ij to destination d is subjected to evaporation (as per 3.3.2. Non-local route management and explicit
Eq. (7)), and (b) when next hop node j to d is no positive reinforcement
longer available (owing to MAC layer errors, route Reactive route discovery is performed by forward
errors, or congestion at j). In the following sections, reactive ants, pf, and backward reactive ants, pb.
we describe the various reinforcement mechanisms Reactive route discovery can be used both at the
at work in ANSI. source of a data packet and at an intermediate node
looking for an alternate route to the destination in
3.3.1. Local route management—reinforcement by the event that previously known routes to the desti-
data packets and the use of neighbor discovery nation have proved ineffective. A route request is
HELLO messages sent by deploying a forward reactive ant pf and
Local route management is made possible by the route reply is sent using a backward reactive
reinforcement due to both movement of data pack- ant, pb. Even though multiple routes can be gath-
ets and an explicit neighbor discovery mechanism. ered by a source sending forward reactive ants (by
These two concepts are illustrated in Fig. 1. allowing the destination to send backward reactive
When a data packet arrives at a node i via a ants in response to all copies of the forward reactive
neighbor node p and is sent to the destination along ants received), we allow the destination to send a
next hop j, both the trail to the previous hop, ip, and backward reactive ant only for the first forward
the trail to the next hop, ij, are reinforced by the SI reactive ant received. This is because we found that
mechanisms at i. in a high traffic/mobility scenario in which a
In addition, nodes in ANSI periodically broad- MANET node has many routes to the destination,
cast a HELLO message. This message can contain packet delivery from source to destination can
a variety of information about the node sending suffer invariably because using several routes will
the message, such as congestion status. In ANSI, spread the traffic over more nodes, and increase
8. ARTICLE IN PRESS
8 S. Rajagopalan, C.-C. Shen / Journal of Systems Architecture xxx (2006) xxx–xxx
the contention in the network. In this case, it seems
like using one route deterministically, while keeping
tabs on the congestion status of neighboring nodes
(which is what ANSI does) is a better approach.3
Regardless, multiple routes are collected owing
to the interaction of the ant information from the
nodes in the network and HELLO beacons, and
are used as and when older routes become defunct.
Also, note that regardless of collecting information
about multiple routes via other mechanisms, ANSI
uses a deterministic choice of next hops when using
Fig. 2. The propagation of the forward reactive ant (shown in
pure MANET nodes (highly capable nodes collect solid arrows) and the return of the backward reactive ant (dashed
multiple routes and use stochastic routing, as we arrows). The rebroadcast from node 2, when received at node 1 is
will see later). This is because we found that stochas- killed immediately to prevent route loops. At each node X the
tic approaches in MANET nodes using ANSI are forward reactive ant enters, it reinforces the path from X to all the
not suited to high data delivery in high traffic other nodes in the nodes-visited-stack. Thus, the forward reactive
ant from S when received at node 4 reinforces the trail 4–2 to
scenarios. both node 2 and node S. On the return path, all nodes in path of
In Fig. 2, consider a node S which needs to route the backward reactive ant reinforce the trails to the paths to all
data packets to D, but does not have a route to D. the nodes in the path leading from the node upstream all the way
Node S buffers the data destined for D and broad- to the destination. Thus, when the backward reactive ant is
casts (over all interfaces) a forward reactive ant, received at S via path 1–3, . . . , D, S will reinforce trail S À 1 for
destinations 1, 3, . . . , D.
pSD (with a nodes visited stack Sf ), intended to dis-
f
cover the route to D. Because there is a good chance
that D has moved, the current implementation of
ANSI sets the number of hops, /f, for the forward 2 and 4), but this is not done for pure MANET
reactive ant (sent from S) to be a few hops larger networks.
than the last known distance of S from D, which In the event that pDS is not received at S within a
b
can be obtained from the routing table at S. If S timeout period, then the value of /f is increased by 2
receives data intended to D after pSD has been
f more hops and the search for the route resumes
broadcast, S buffers the data. When D receives again. The process of route discovery is continued
pSD , D copies the nodes visited stack, Sf , into a
f again if a route is not found after the second try.
new backward reactive ant, pDS , and kills pSD . D
b f ANSI retries twice for a route to destination.
then sends pDS to S. pDS is not broadcast, it just
b b To control the amount of MAC layer usage at a
backtracks to the source S by using the nodes vis- node X, a scheduled HELLO packet is broadcast at
ited stack Sb in pDS . The ant, pDS , when visiting a
b b X only if the last broadcast forward reactive ant
node X along the path to S positively reinforces was sent before the last HELLO message.
the route to all nodes v 2 Sb upstream from X to
D, and adds an entry in AX to D via the next hop 3.3.3. Route errors, and negative reinforcement
immediately upstream (in the path from S to D). Route errors occur at a node X when X is unable
An intermediate node thereby knows what next to provide a route for the destination D owing to
hop to use to route to D. In this way, backward non-availability of a routing table entry at X or
reactive ants perform explicit positive reinforcement due to the non-availability of the next hop suggested
of routes to destination D. When S receives pDS b by the routing table entry at X. When a route error
from D, S sends the buffered packets intended for occurs at a node X in a network running ANSI, X
D over the newly discovered route and flushes S’s first buffers the packet which X needs to forward
buffer. Note that multiple paths may be readily col- and then sends a forward reactive ant to find the
lected (for example, by sending another backward destination D. If X happens to be an intermediate
reactive ant for the ant proceeding to D via nodes node, in addition to sending a forward reactive
ant, X also sends a route error back to the source
3
Stochastic approaches to routing in pure MANET networks
S of the packet. The packets buffered at X are
is an effective approach when the mobility and traffic in the relayed across the network after a backward reac-
network are low. tive ant from D reaches X.
9. ARTICLE IN PRESS
S. Rajagopalan, C.-C. Shen / Journal of Systems Architecture xxx (2006) xxx–xxx 9
In addition, when a route error is received at an
intermediate node between X and S, the node
explicitly invalidates the routing table entries to D.
The packets received at X before the route error is
received at S are X’s responsibility (to forward),
but the packets generated after the time when the
route error is received at S from X are S’s responsi-
bility—S generates a forward reactive ant to find the
route to D.
3.3.4. Proactive routing within highly capable sections
of the network Fig. 3. In a hybrid network (nodes 1–9 are part of a highly
As mentioned in Section 3.1, nodes belonging to capable network in this case, and are connected by the grid
shown), nodes belonging to more capable infrastructure are able
non-mobile, highly capable infrastructure, such as to perform stochastic routing. In this figure, two possible paths
cellular networks engage in proactive routing as well that 1 can take to route to D are one via nodes 1 ! 4 ! 7 !
as reactive routing because these nodes are not con- 8 ! 9 (P1) and another via nodes 1 ! 2 ! 5 ! 6 ! 9 (P2).
cerned about topological fluctuations. These nodes
also maintain a list of mobile nodes which are acces-
sible from each other, thus assisting the reactive the changes in the network. Hence, a choice for a,
routing process within the mobile nodes as and b, and c should be made carefully to allow for
when possible. Nodes in non-mobile, highly capable responsiveness of the system. Using insights from
infrastructure send proactive ants periodically to all our preliminary results, we arrived at a value of
the other highly capable nodes they are connected a = b = c = 2, and these are the values we use in
to. Proactive ants are not returned like forward our implementation.
reactive ants, and they reinforce the route to the
proactive ant sender along the path the proactive 4. Simulation results
ant takes. Proactive ants, apart from carrying a
nodes-visited-stack for gathering information about ANSI was simulated in QualNet (Version 3.7),
the nodes that were visited, are fixed in hop length and the performance of ANSI was compared with
and also carry a data structure for indicating the a popular routing protocol, AODV [1], for the same
mobile nodes which are accessible from the proac- network and load characteristics. We chose to com-
tive ant sender. These nodes engage in proactive pare ANSI with AODV because AODV has been
route collecting activity using all their interfaces, shown to perform well in a vast majority of ad
and so are able to combine routes found via differ- hoc network scenarios. In addition, AODV also
ent interfaces effectively during the routing process. works on hybrid ad hoc networks. Our work here
Because nodes belonging to a highly capable net- is an extension of our earlier work [19] which only
work need not be concerned about the issues due to tested ANSI under UDP loads over a pure
mobility, these nodes are able to effectively utilize MANET. As we mentioned earlier, ANSI functions
the benefits due to stochastic routing (see Fig. 3). as a purely reactive protocol in a pure MANET
As mentioned before, ant-based routing naturally environment.
lends itself to stochastic routing because multiple
routes are found and maintained. 4.1. Simulation and network model
3.3.5. Driving the routing process via more desirable 4.1.1. ANSI parameters
nodes The current implementation of ANSI used
By choosing higher values for a, b, and c, the a = b = c = 2. In both AODV and ANSI, the reac-
process of next hop selection in ANSI favors the tive route recovery is retried twice, and for ANSI,
next hops with higher values for s, g, and w, respec- the last try uses /f = 15. For the first two trials in
tively. However, by choosing values which are too ANSI, /f is determined according to the informa-
high, the route selection is too skewed towards the tion available about the unknown destination: if
best next hops and it becomes very difficult for the the destination had a valid entry in the routing table
SI mechanisms at the nodes to respond quickly to earlier, then /f is set to one more than the earlier
10. ARTICLE IN PRESS
10 S. Rajagopalan, C.-C. Shen / Journal of Systems Architecture xxx (2006) xxx–xxx
a b
Fig. 4. Hybrid network topologies used for Experiments 1–3: (a) The hybrid network topology for Experiments 1 and 2. (b) The hybrid
network topology for Experiment 3.
number of hops to the destination. Otherwise, (thus, one stream will be an ‘‘internal’’ stream).
/f = 5. The evaporation constant, c, used in Eq. There are thus, altogether, 16 traffic streams in this
(7) is 15 s. In hybrid networks, the nodes which experiment. Each of these streams send 512-byte
are part of the high-speed Ethernet (see Section packets at a uniform rate of 1–20 packets/s.
4.1.2) used a proactive route update interval of 10 s. In Experiment 3, we studied the performance of
ANSI and AODV in a larger hybrid network con-
4.1.2. Network and application parameters sisting of 360 pure MANET nodes spread over 9
We performed five experiments, in which we mobile regions uniformly located in a 5000 m ·
studied the performance of ANSI and AODV with 5000 m terrain, each of size 1000 m · 1000 m and
increasing traffic and increasing number of nodes serviced by one highly capable, immobile node
of both UDP and TCP flows in both hybrid and (nodes 361–369) located at the center of each mobile
pure MANETs. In all these experiments, the source region. The highly capable nodes are all connected
and destination are chosen randomly and are pair- via a 100 Mbps Ethernet link. The topology of our
wise-distinct for each trial. experiment is shown in Fig. 4(b). Each highly capa-
In Experiments 1 and 2, we studied the perfor- ble node has both Ethernet interfaces and an 802.11
mance of ANSI vs. AODV in a hybrid network, interface. The size of the data packets sent was
for both UDP (Experiment 1) and TCP (Experiment 512 bytes. Six UDP streams are randomly gener-
2) flows. In these experiments, the non-mobile nodes ated, with the following profile of the source–desti-
are connected to each other over a 100 Mbps Ether- nation pairs: (a) regions 1–4, (b) regions 1–7, (c)
net link. Fig. 4(a) shows the simulation topology. regions 8–5, (d) regions 8–2, (e) regions 3–6, and
The size of the entire terrain is 2000 m · 2000 m. (f) regions 3–9. The data sources generated packets
Inside this terrain, there are four MANET at the uniform rate of 2 to 20 packets/s in steps of
‘‘regions’’, each of which contain 20 MANET nodes 2 packets/s.
inside a terrain of size 500 m · 500 m, and ‘‘ser- In Experiment 4, we studied the effect of increas-
viced’’ by one highly capable, immobile node (nodes ing TCP traffic in a pure MANET network. In this
81–84) located in the center of the mobile region. experiment, 50 nodes were placed uniformly in a
This highly capable node, located in the center of network of size 1100 m · 1100 m. This maintains a
each of the regions, manages both an Ethernet inter- node density4 of 8.15 mÀ2, which, according to
face and an 802.11 interface, and is connected to the [20], is sparse for a network with mobile nodes.
others by another highly capable node, node 85, The experiment simulates 25 streams of TCP traffic
which has 4 Ethernet interfaces. Note that MANET sending 64-byte packets at a uniform packet rate
nodes within a region are not able to communicate varying from 1 to 20 packets/s.
with MANET nodes of other regions directly (the
closest they can get is around 353 m, which is beyond
the transmission range of the MANET nodes).
Four streams are chosen for each region, with 4
Node density is defined as the number of nodes in an area
one stream headed towards each of the regions covered by the transmission range of a node.
11. ARTICLE IN PRESS
S. Rajagopalan, C.-C. Shen / Journal of Systems Architecture xxx (2006) xxx–xxx 11
In Experiment 5, we studied the performance of aged over the number of source–destination
ANSI and AODV under UDP loads in a pure pairs. For TCP flows, the above described quan-
MANET environment with an increasing number tity is the measured packet delivery ratio. The
of nodes. The number of nodes was varied from actual packet delivery for TCP flows is calculated
50 to 250 and exactly half the number of nodes were using the expected number of packets that should
data sources. The terrain size was such that the node be sent at the application layer at the data
density was constant at 8.15 mÀ2 (for example, for sources. UDP does not perform congestion-con-
50 nodes, the terrain size was 1100 m · 1100 m). trol so the expected and measured number of
The data sources generated one 64-byte packet a packets sent at the application layer of the data
second to be sent to the data sink. source are the same.
In all the experiments, the MANET nodes were 2. (End-to-end metric 2) End-to-end delay: measured
uniformly distributed initially in the terrain and as the average delay in sending packets from
the mobile nodes moved as per the Random Way- source to destination and averaged over the num-
point Model with a minimum speed of 0.001 m/s, ber of source–destination pairs.
maximum speed of 20 m/s, with a pause time of 3. (End-to-end metric 3) Delay jitter: measured as
10 s. In hybrid networks (Experiments 1–3), the the average variance of the interarrival times at
mobile nodes were restricted to move only within the destinations and averaged over the number
their region (bounded by a 500 m · 500 m terrain of source–destination pairs.
for Experiments 1 and 2 and a 1000 m · 1000 m ter- 4. (End-to-end metric 4) Number of packets sent by
rain for Experiment 3). The MANET nodes in the Super application sender: measured as the total
experiments used one 802.11 interface with omnidi- number of packets which are actually sent by
rectional antennas and a transmission range of Super Application senders. For Super Applica-
250 m at the physical layer and 802.11DCF at the tion using TCP, this number depends on how
MAC layer. The link bandwidth for the mobile long the TCP connection lasts.
nodes using 802.11 was 2 Mbps. In addition to using 5. (End-to-end metric 5) Variation of the congestion
802.11, the non-mobile nodes also used Ethernet window of a sender: measured as the TCP conges-
with a capacity of 100 Mbps. The simulations used tion window (snd_cwnd) at one sender for one
a two-ray pathloss model and no propagation fad- flow for one trial as it varies with simulation time.
ing model was assumed. The application used was 6. (Network-wide metric 1) Total number of route
CBR, and sources and destinations were pairwise errors initiated: is the total number of route
distinct and chosen randomly. Both TCP and errors generated in the network.
UDP-based CBR flows were studied. Super applica- 7. (Network-wide metric 2) Total number of 802.11
tion was used for generating a reliable CBR traffic DCF MAC layer unicasts sent: is the total num-
stream using TCP (regular CBR application uses ber of all (successful) 802.11DCF unicast trans-
UDP). missions sent in the network. For AODV, this
All experiments were run for a simulated time of measures the total number of data packets,
5 min and all sources started sending packets at RREP and RERR sent out at the 802.11DCF
exactly 40 s into the simulation and ended data gen- interface. For ANSI, this measures the total
eration at exactly 260 s. TCP-LITE, used for the number of data packets, backward reactive ants,
TCP flows in our experiments, is a variant of and RERRs sent at the 802.11 interface.
TCP-RENO, and used an MSS of 512 bytes, maxi- 8. (Network-wide metric 3) Total number of 802.11
mum send/receive buffer of 16384 bytes each, and DCF MAC layer broadcasts sent: is the total
delayed ACKs. number of all 802.11DCF broadcasts sent by all
We studied the following end-to-end and net- nodes in the network. For AODV, this measures
work-wide characteristics: the total number of RREQ and Hello packets
sent at the 802.11DCF interfaces, and for ANSI,
1. (End-to-end metric 1) Packet delivery fraction: this measures the total number of forward reac-
measured at the application layer as the ratio of tive ants, proactive ants and the Hello packets
the total number of packets which were received sent at the 802.11 interfaces.
(at the application layer) at the data sinks to the
total number of packets that were sent from the We do not report end-to-end delay and delay jit-
data sources (at the application layer), and aver- ter for TCP flows as these metrics are typically not
12. ARTICLE IN PRESS
12 S. Rajagopalan, C.-C. Shen / Journal of Systems Architecture xxx (2006) xxx–xxx
reported for TCP flows, because of the fact that AODV send a comparable number of MAC uni-
TCP has to deal with out-of-order deliveries casts. ANSI sends fewer MAC broadcasts when
and the large delays (as compared to UDP flows) the packet rate is low to moderate, but as the packet
owing to reliability and congestion-control mecha- rate increases, ANSI sends more MAC broadcasts.
nisms. The reason why ANSI performs better than
We analyzed the results from the above experi- AODV—delivering more packets with better met-
ments and show them using graphs with 95% confi- rics such as delay, jitter and number of route
dence intervals of the measured values. errors—is because ANSI manages the local network
information better than AODV does, and performs
4.2. Simulation results congestion-aware routing. This is why ANSI shows
lower route errors as compared to AODV (see
4.2.1. Experiment 1: Hybrid network—effect of Fig. 5(d)). Owing to the above reasons, routes break
increasing the UDP packet rate less often and result in fewer route request opera-
Fig. 5 shows the results for the performance of tions in ANSI as compared to AODV. When routes
ANSI vs. AODV over a hybrid network using do break in ANSI, they are managed by the proto-
UDP flows. We see that ANSI consistently outper- col mechanisms locally rather than a network-wide
forms AODV in terms of packet delivery, delay, jit- flooding. This in turn results in lower congestion
ter and number of RERR initiated. ANSI and at the nodes. This is why, even though ANSI shows
Packet delivery at the CBR layer (%)
1 0.3
AODV
0.99 0.25
End-to end delay (s)
ANSI
0.2
0.98
0.15
0.97
0.1
AODV
0.96 0.05
ANSI
0.95 0
0 5 10 15 20 0 5 10 15 20
a Packet rate (pkts/s) b Packet rate (pkts/s)
600
Total number of RERR initiated
AODV AODV
500
ANSI ANSI
0.1
Delay jitter (s)
400
300
0.05 200
100
0 0
0 5 10 15 20 0 5 10 15 20
c Packet rate (pkts/s) d Packet rate (pkts/s)
4 4
x 10 x 10
15 5
802.11DCF,Broadcasts sent
802.11DCF,Unicasts sent
4
10
3
2
5
AODV 1 AODV
ANSI ANSI
0 0
0 5 10 15 20 0 5 10 15 20
e Packet rate (pkts/s) f Packet rate (pkts/s)
Fig. 5. Experiment 1: Performance studies of ANSI vs. AODV in a hybrid network with UDP flows: (a) Packet delivery ratio, (b) end-to-
end delay, (c) delay jitter, (d) number of RERR initiated, (e) 802.11DCF, Unicasts sent and (f) 802.11DCF, Broadcasts sent.
13. ARTICLE IN PRESS
S. Rajagopalan, C.-C. Shen / Journal of Systems Architecture xxx (2006) xxx–xxx 13
larger MAC broadcasts in higher packet rates, it As packet rate increases, for both AODV and
still shows delays and jitter lower than that for ANSI, the mean time before links break owing to
AODV. The higher number of broadcasts when node mobility is still the same, but because the of
the packet rate increases is because of the conges- the use of highly capable nodes (which are within
tion-aware properties of ANSI, which allow it to 2 hops away for any MANET node), the percentage
drop badly congested routes and look for new ones. of packets delivered with lower variation in end-to-
This, while delivering packets more quickly and end delay increases (in comparison to the number of
smoothly, obviously makes ANSI incur more route packets delivered at higher variations in end-to-end
discovery overheads, which is what we see in terms delay) at both AODV and ANSI, thus bringing the
of larger MAC broadcast overheads. The new, con- overall variation down. This is why we see a
gestion-free (or low congestion) routes are then used decrease in delay jitter as packet rate increases for
to deliver more packets in ANSI. Note that AODV, both ANSI and AODV.
does not show an appreciable increase in the num-
ber of MAC broadcasts as the packet rate increases 4.2.2. Experiment 2: Hybrid network—effect of
because it does not perform congestion-aware rout- increasing the TCP packet rate
ing, but owing to this, the performance of AODV Fig. 6 shows the results for Experiment 2. For
degrades. The fact that the number of ANSI’s TCP flows, we see that ANSI’s measured and actual
MAC unicasts are comparable to that of AODV packet delivery ratio is higher than the same metrics
(in the context of better performance metrics), along for AODV. We also see that for AODV, the mea-
with its fewer route errors is a clear indication of the sured packet delivery ratio improves as the packet
fact that ANSI is engaged in providing/finding bet- rate increases, but the actual packet delivery ratio
ter routes as compared to AODV. decreases. ANSI’s actual packet delivery is nearly
The reason why delay jitter decreases with 5–10% more than AODV’s actual packet delivery
increasing packet rate in both ANSI and AODV ratio. ANSI also sends more packets during the sim-
(see Fig. 5(b) and (c)) is as follows: Delay jitter is ulation as compared to AODV—we see that the
a measure of the variation of interarrival times at number of packets which ANSI sends is very close
the destination. Thus, if end-to-end delay measured to the number expected to be sent.5 In terms of
at the destination varies very little, then delay jitter the effect of the routing protocol on TCP, the con-
is bound to be low. ANSI, being congestion-aware, gestion window for the output queue at node 53
chooses congestion-free routes and delivers packets (for packet rate 1 packets/s, sent from node 53 to
at the destination with little variation in end-to- node 48) shows steady growth, while AODV’s con-
end delay. AODV, because it is not congestion- gestion window (for the same stream, output queue
aware, delivers packets along congested routes, at node 53) shows substantially slower growth.
which results in higher end-to-end delays because ANSI, as before, shows a lot fewer route errors
a node running AODV does not react to congestion (see Fig. 6(d)). ANSI shows more MAC unicast
until a congested node along the path is no longer traffic as compared to AODV. Though ANSI shows
able to receive or transmit packets. Thus, for a sin- lower MAC broadcast traffic when the packet rate is
gle stream of UDP traffic from one source to desti- low, it shows more MAC broadcast overheads when
nation in AODV, the destination first experiences the packet rate increases.
low variation in end-to-end delay, but thereafter, The reason why ANSI performs better (with 5–10%
the path becomes more congested and the variation higher actual packet delivery ratio) than AODV under
in end-to-end delay progressively increases until the TCP loads is because of the congestion-aware routing
path breaks. AODV then engages in route discovery in ANSI. Owing to this property, ANSI is able to sup-
and finds a congestion-free path, and once again the ply congestion-free routes which allow for the smooth
measurement of end-to-end delay at the destination passage of ACKs back to the data source, allowing
shows low variation until the new path becomes TCP operations to perform smoothly.
congested again. Statistically, the value of delay jit- We would like to draw attention to the graphs
ter depends on the percentage of the packets that showing the measured packet delivery ratio in
are delivered at the destination with low variation Fig. 6(a). These results for measured packet delivery
in end-to-end delay, and so if a higher percentage
of packets are delivered with a higher variation, 5
This amount is 16 · 220 · x = 3520x packets total (x is the
the jitter is bound to be larger. packet rate), and indicated by the straight line graph in Fig. 6(b).
14. ARTICLE IN PRESS
14 S. Rajagopalan, C.-C. Shen / Journal of Systems Architecture xxx (2006) xxx–xxx
4
Packet delivery at the app. layer (%)
x 10
1
Packets sent by the app. layer
6
AODV(m)
0.95
AODV(a)
ANSI(m) 4
0.9 ANSI(a)
AODV
2 ANSI
0.85
Ideal
0.8 0
0 5 10 15 20 0 5 10 15 20
a Packet rate (pkts/s) b Packet rate (pkts/s)
250
Total number of RERR initiated
ANSI
Congestion window (bytes)
15000 AODV
200
10000 150
100
5000
AODV
50
ANSI
0
0 100 200 300 0
0 5 10 15 20
c Simulation time (s) d Packet rate (pkts/s)
5 4
x 10 x 10
3 4
802.11DCF,Broadcasts sent
802.11DCF,Unicasts sent
2.5
3
2
1.5 2
1
AODV 1 AODV
0.5
ANSI ANSI
0 0
0 5 10 15 20 0 5 10 15 20
e Packet rate (pkts/s) f Packet rate (pkts/s)
Fig. 6. Experiment 2: Performance studies of ANSI vs. AODV in a hybrid network with TCP flows: (a) measured (m) and actual
(a) packet delivery ratio, (b) packets sent by the Super Application Layer, (c) congestion window for TCP output at node 53 for 1 pkt/s,
(d) number of RERR initiated, (e) 802.11DCF, Unicasts sent and (f) 802.11DCF, Broadcasts sent.
ratio are counter-intuitive. While the measured that we had fixed the TCP send buffer to be
packet delivery ratio of AODV increases with 16,384 bytes, and the congestion window cannot
packet rate, we note that the percentage of packets grow beyond this size. In this figure, we see how
sent to the data sink increasingly decreases. Thus, Super application works TCP when sending CBR
the actual packet delivery ratio, measured as the traffic. Note that this is traffic inside a mobile region
percentage of packets that are received to the per- (both node 53 and node 48 are inside the same
centage of packets that are expected to be sent (in mobile region as per Fig. 4(a)). TCP, when working
this case x · 16 · 220 = 3520x, where x is the packet on top of ANSI, is able to increase the congestion
rate), actually decreases. So, the traditional packet window as per congestion avoidance algorithms,
delivery ratio metrics, defined as the ratio of the but in AODV, congestion avoidance is quickly
number of application layer packets delivered to thwarted by congestion occurring along the path
the number of application packets sent, is actually from node 53 to node 48, which is why the TCP stack
a misleading metric to measure when studying at node 53 shows fast recovery behavior [21] for the
MANET performance under TCP loads. TCP output queue. This is the case owing to losing a
The behavior of ANSI and AODV under TCP lot of ACKs in AODV. Indeed, we see that the con-
loads can be summarized clearly by Fig. 6(c). Note gestion window in AODV does not grow/change
15. ARTICLE IN PRESS
S. Rajagopalan, C.-C. Shen / Journal of Systems Architecture xxx (2006) xxx–xxx 15
after a certain point into the simulation (around More packets are sent by ANSI at the Super
200 s) for AODV. Whereas, for ANSI, we see a application layer as a result of larger congestion
‘‘healthy’’ growth of the congestion window, con- windows, and subsequently, more MAC unicasts
trolled by the congestion-avoiding sender (linear are sent for these packets, which is why we see the
growth of congestion window) rather than being number of MAC unicasts for ANSI is more. More
controlled by congestion elsewhere in the network. MAC broadcasts are sent in ANSI as a response
This behavior for TCP running over ANSI to finding newer routes which are less congested.
results from ANSI’s congestion awareness, which As before, AODV does not respond to congestion,
constantly maintains routes with low congestion and so it shows only a small increase in the number
and chooses them in favor of the ones with higher of MAC broadcasts as the packet rate increases.
congestion. This permits TCP running over ANSI
to receive ACKs more frequently and regularly than 4.3. Experiment 3: Large hybrid network—effect
in the AODV case, where losing ACKs causes fast of increasing UDP packet rate
recovery behavior. AODV, not being congestion
aware, chooses congested routes frequently because Fig. 7 shows the results for the performance of
it has no way of knowing which routes are con- ANSI and AODV in a larger hybrid network. As
gested and which ones are not, making the passage we can see, the results are similar to the results of
of ACKs more difficult. Experiment 1, shown in Fig. 5. We also see ANSI’s
Packet delivery at the CBR layer (%)
1
AODV
AODV 0.4
0.95 ANSI
End-to-end delay (s)
ANSI
0.9 0.3
0.85
0.2
0.8
0.1
0.75
0.7 0
5 10 15 20 0 5 10 15 20
a Packet rate (pkts/s) b Packet rate (pkts/s)
0.4
Total number of RERR initiated
AODV 1500 AODV
0.3 ANSI ANSI
Delay jitter (s)
1000
0.2
500
0.1
0 0
0 5 10 15 20 5 10 15 20
c Packet rate (pkts/s) d Packet rate (pkts/s)
4 5
x 10 x 10
10
802.11DCF,Broadcasts sent
3
802.11DCF,Unicasts sent
8
2.5
6 2
1.5
4
1
2 AODV AODV
ANSI 0.5 ANSI
0 0
0 5 10 15 20 0 5 10 15 20
e Packet rate (pkts/s) f Packet rate (pkts/s)
Fig. 7. Experiment 3: Performance studies of ANSI vs. AODV in a (larger) hybrid network with UDP flows: (a) packet delivery ratio,
(b) end-to-end delay, (c) delay jitter, (d) number of RERR initiated, (e) 802.11DCF, Unicasts sent and (f) 802.11DCF, Broadcasts sent.