Computer Science Technical Report Archive

USC Computer Science Technical Reports, no. 789 (2003)

(USC DC Other)

USC Computer Science Technical Reports, no. 789 (2003)

PDF

Download a page range

Download transcript

Copy asset link

Request this asset

Transcript (if available)

Content 1
the IMPORTANT framework for analyzing the
Impact of Mobility on Performance of RouTing
protocols for Adhoc NeTworks
Fan Bai, Narayanan Sadagopan, Ahmed Helmy
Department of Electrical Engineering
Department of Computer Science
University of Southern California
April 30, 2003 DRAFT
2
Abstract
A Mobile Ad hoc Network (MANET) is a collection of wireless mobile nodes forming a temporary network
without using any existing infrastructure. Since not many MANETs are currently deployed, research in this area is
mostly simulation based. Random Waypoint is the commonly used mobility model in these simulations. Random
Waypoint is a simple model that may be applicable to some scenarios. However, we believe that it is not sufficient
to capture some important mobility characteristics of scenarios in which MANETs may be deployed. Our framework
aims to evaluate the impact of different mobility models on the performance of MANET routing protocols. We
propose various protocol independent metrics to capture interesting mobility characteristics, including spatial and
temporal dependence and geographic restrictions. In addition, a rich set of parameterized mobility models is introduced
including Random Waypoint, Group Mobility, Freeway and Manhattan models. Based on these models several ’test-
suite’ scenarios are chosen carefully to span the metric space. We demonstrate the utility of our test-suite by evaluating
various MANET routing protocols, including DSR, AODV and DSDV . Our results show that the protocol performance
may vary drastically across mobility models and performance rankings of protocols may vary with the mobility
models used. This effect can be explained by the interaction of the mobility characteristics with the connectivity
graph properties. Finally, we attempt to decompose the reactive routing protocols into mechanistic “building blocks”
to gain a deeper insight into the performance variations across protocols in the face of mobility.
I. INTRODUCTION
A Mobile Ad hoc NETwork (MANET) is a collection of wireless nodes communicating with each other in
the absence of any infrastructure. Classrooms, battlefields and disaster relief activities are a few scenarios where
MANETs can be used. MANET research is gaining ground due to the ubiquity of small, inexpensive wireless
communicating devices. Since, not many MANETs have been deployed, most of this research is simulation based.
These simulations have several parameters including the mobility model and the communicating traffic pattern.
In this paper, we focus on the impact of mobility models on the performance of MANET routing protocols. We
acknowledge that the communicating traffic pattern also has a significant impact on the routing protocol performance
and merits a study on its own. However, as in most studies in this area, in order to isolate the effect of mobility,
we fix the communicating traffic pattern to consist of randomly chosen source-destination pairs with long enough
session times.
Mobility pattern, in many previous studies was assumed to be Random Waypoint. In the current network simulator
(ns-2) distribution, the implementation of this mobility model is as follows: at every instant, a node randomly chooses
a destination and moves towards it with a velocity chosen uniformly randomly from [0;V
max
], where V
max
is the
maximum allowable velocity for every mobile node [1]. Most of the simulations using the Random Waypoint
model are based on this standard implementation. For the rest of the paper, we refer to this basic implementation
as the Random Waypoint model.
In the future, MANETs are expected to be deployed in myriads of scenarios having complex node mobility and
connectivity dynamics. For example, in a MANET on a battlefield, the movement of the soldiers will be influenced
by the commander. In a city-wide MANET, the node movement is restricted by obstacles or maps. The node
mobility characteristics are very application specific. Widely varying mobility characteristics are expected to have
April 30, 2003 DRAFT
3
a significant impact on the performance of the routing protocols like DSR [5], DSDV [6] and AODV [7]. Random
Waypoint is a well-designed and commonly used mobility model, but we find it is insufficient to capture those
characteristics, such as:
1) Spatial dependence of movement among nodes.
2) Temporal Dependence of movement of a node over time.
3) Existence of barriers or obstacles constraining mobility.
In this study, we focus on the impact of the above mentioned mobility characteristics on protocol performance.
While doing so, we propose a generic framework to systematically analyze the impact of mobility on the performance
of routing protocols for MANETs. This analysis attempts to answer the following questions:
1) Whether and to what degree mobility affects routing protocol performance?
2) If the answer to 1 is yes, why?
3) If the answer to 1 is yes, how?
To answer Whether, the framework evaluates the performance of these routing protocols over different mobility
patterns that capture some of the characteristics listed above. The mobility models used in our study include
the Random Waypoint, Group Mobility [8], Freeway and Manhattan. To answer Why, we propose some protocol
independent metrics such as mobility metrics and connectivity graph metrics. Mobility metrics aim to capture
some of the aforementioned mobility characteristics. Connectivity graph metrics aim to study the effect of different
mobility patterns on the connectivity graph of the mobile nodes. It has also been observed in previous studies that
under a given mobility pattern, routing protocols like DSR, DSDV and AODV perform differently [9] [10] [11].
This is possibly because each protocol differs in the basic mechanisms or “building blocks” it uses. For example,
DSR uses route discovery, while DSDV uses periodic updates. To answer How, we want to investigate the effect
of mobility on some of these “building blocks” and how they impact the protocol performance as a “whole”.
In order to conduct our research and answer the above questions systematically, we propose a framework for
analyzing the Impact of Mobility on the Performance Of RouTing protocols in Adhoc NeTworks (IMPORTANT).
Through this framework we illustrate how modeling mobility is important in affecting routing performance and
understanding the mechanism of ad hoc routing protocols. As shown in Fig.1, our framework focuses on the
following aspects: mobility models, the metrics for mobility and connectivity graph characteristics, the potential
relationship between mobility and routing performance and the analysis of impact of mobility on building blocks
of ad hoc routing protocols.
The rest of this paper is organized as follows. Section II gives a brief description of the related work and elaborates
our contribution. Section III discusses some limitations of the Random Waypoint model and motivates part of our
framework. Section IV presents our proposed metrics to capture characteristics of mobility and the connectivity
graph between the mobile nodes. Section V describes the mobility models used and introduces two new models, the
Freeway mobility model and the Manhattan mobility model. Results of our simulation experiments are presented
and discussed in Section VI. The analysis of the impact of mobility on protocol building blocks is discussed in
April 30, 2003 DRAFT
4
Mobility
Models
Mobility
Metrics
Connectivity
Graph
Connectivity
Metrics
Performance
Metrics
Routing
Protocol
Performance
Random Waypoint
Group Mobility
Freeway Mobility
Manhattan Mobility
DSR
AODV
DSDV
Relative Speed
Spatial Dependence
Link Duration
Path Duration
Throughput
Overhead
Building
Block
Analysis
Flooding
Caching
Error Detection
Error Handling
Error Notification
Fig. 1. IMPORTANT Framework
Section VII. Finally, our conclusions from this study and planned future work are listed in section VIII.
II. RELATED WORK
Extensive research has been done in modeling mobility for MANETs. In this section, we mainly focus on
experimental research in this area. This research can be broadly classified as follows based on the methodology
used:
A. Random Waypoint Based Performance Comparisons
Much of the initial research was based on using Random Waypoint as the underlying mobility model and CBR
traffic consisting of randomly chosen source destination pairs as the traffic pattern. Routing protocols like DSR [5],
DSDV [6], AODV [7] and TORA [12] were mainly evaluated based on the following metrics: packet delivery ratio
(ratio of the number of packets received to the number of packets sent) and routing overhead (number of routing
control packets sent). [9] concluded that on-demand protocols such as DSR and AODV performed better than table
driven ones such as DSDV at high mobility rates, while DSDV performed quite well at low mobility rates. [10]
performed a comparison study of the two on-demand routing protocols: DSR and AODV , using the performance
metrics of packet delivery ratio and end to end delay. It observed that DSR outperforms AODV in less demanding
situations, while AODV outperforms DSR at heavy traffic load and high mobility. However, the routing overhead of
DSR was found to be lesser than that of AODV . In the above studies, focus was given on performance evaluation,
while parameters investigated in the mobility model were change of maximum velocity and pause time. In our
work, however, we design our test suites very carefully to pick scenarios that span a much larger set of mobility
characteristics. Not only do we use Random Waypoint but also other mobility models such as RPGM [8], Freeway
and Manhattan in our evaluation of the performance of routing protocols.
B. Scenario Based Performance Comparisons
Random Waypoint is a simple model that is easy to analyze and implement. This has probably been the main
reason for the widespread use of this model for simulations. Realizing that Random Waypoint is too general a
April 30, 2003 DRAFT
5
model, recent research has started focusing on alternative mobility models and protocol independent metrics to
characterize them. [13] conducted a scenario based performance analysis of the MANET protocols. It proposed
models for a few “realistic” scenarios such as a conference, event coverage and disaster relief. To differentiate
between scenarios used, the study introduced the relative motion of the mobile nodes as a mobility metric. Their
conclusions about the performance of proactive and reactive protocols were similar to [9]. [11] used a mobility
model in which each node computes its next position based on a probability distribution. This model does not allow
significant changes in direction between successive instants. It concluded that proactive protocols perform better
than reactive ones in terms of packet delivery ratio and end-to-end delay. However, reactive protocols were seen
to incur a lower routing overhead. [8] introduced the Reference Point Group Mobility (RPGM) model, which is
one of the mobility models used in this study. Rate of link changes was used to characterize a few group mobility
patterns as well as Random Waypoint. It observed that the rate of link change for Random Waypoint was higher
than that for RPGM. From experiments, it observed that protocols like AODV , DSDV and HSR [14] perform
worse with Random Waypoint than with RPGM. Thus, it concluded that mobility models do matter and it is not
sufficient to simulate protocols with only the “random walk” like models. [15] proposed a mobility framework that
consisted of a Mobility Vector Model which can be used to generate “realistic” movement patterns used in several
varied applications. It proposed the Displacement Measure that is a normalization of the actual distance traveled by
the geographic displacement as a metric to evaluate the different movement patterns including those generated by
Random Waypoint, Random Walk, RPGM and Mobility Vector models. By experiments, it observed that Random
Waypoint and Random Walk produced higher Displacement Measure as compared to the Mobility Vector model. It
studied the effect of transmission range on throughput across different mobility models and concluded that as the
transmission range is increased, the rate of link changes decreased and the throughput for all protocols increased.
However, the link change rate does not seem to vary greatly across the different mobility models. As far as routing
overhead was concerned, Mobility Vector was seen to produce a worse overhead than Random Waypoint. Our study
is also framework based. However, we do not aim to provide a generic mobility model from which all “realistic”
mobility patterns can be derived. Rather, our framework aims at systematically studying the effect of mobility per
se on performance of MANET routing protocols. The contributions of our proposed framework are three fold:
1) Focus on mobility characteristics such as spatial dependence, geographic restrictions and temporal dependence.
Define mobility metrics that capture these characteristics. Choose mobility models that span the metric space
and use them to evaluate the performance of routing protocols.
2) Define connectivity graph metrics. Study the interaction of mobility metrics and connectivity graph metrics
and its effect on protocol performance.
3) Analyze the reasons for the differences in protocol performance as a “whole” by investigating the effect of
mobility on “parts” that build the protocol.
April 30, 2003 DRAFT
6
III. LIMITATIONS OF RANDOM WAYPOINT
Random Waypoint model was introduced in [9] and is among the most commonly used mobility models in the
MANET research community. In this model, at every instant, each mobile node chooses a random destination and
moves towards it with a speed uniformly distributed in [0;V
max
], where V
max
is the maximum allowable speed for
a node. After reaching the destination, the node stops for a duration defined by the “pause time” parameter. After
this duration, it again chooses a random destination and repeats the whole process again until the simulation ends.
The Random Waypoint model is widely accepted mainly due to its simplicity of implementation and analysis.
However, we observe that the basic Random Waypoint model as used in most of the simulations is insufficient to
capture the following mobility characteristics:
1) Temporal dependency: Due to physical constraints of the mobile entity itself, the velocity of mobile node
will change continuously and gently instead of abruptly, i.e. the current velocity is dependent on the previous
velocity. However, intuitively, the velocities at two different time slots are independent in the Random
Waypoint model.
2) Spatial dependency: The movement pattern of a mobile node may be influenced by and correlated with
nodes in its neighborhood. In Random Waypoint, each mobile node moves independently of others.
3) Geographic restrictions: In many cases, the movement of a mobile node may be restricted along the street
or a freeway. A geographic map may define these boundaries.
In our study, we focus on the above-mentioned characteristics. In the next section, we formally define metrics to
capture some of these characteristics.
IV. METRICS
To quantitatively and qualitatively analyze the impact of mobility on routing protocol performance, we make use
of several protocol independent metrics and protocol performance metrics. The protocol independent metrics attempt
to extract the characteristics of mobility and the connectivity graph between the mobile nodes. These metrics are
then used to explain the impact of mobility on the protocol performance metrics. Those metrics can be broadly
classified as:
1) Mobility Metrics.
2) Connectivity Graph Metrics.
3) Protocol Performance Metrics.
A. Terminology
Before formally defining the metrics, we introduce some basic terminology that will be used later in the paper:
1)
~
V
i
(t): Velocity vector of node i at time t.
2) j
~
V
i
(t)j: Speed of node i at time t.
3) i
(t): Angle made by
~
V
i
(t) at time t with the X-axis.
April 30, 2003 DRAFT
7
4) ~ a
i
(t): Acceleration vector of node i at time t.
5) x
i
(t): X co-ordinate of node i at time t.
6) y
i
(t): Y co-ordinate of node i at time t.
7) D
i;j
(t): Euclidean Distance between nodes i and j at time t.
8) RD ( ~ a(t);
~
b(t
0
)): Relative Direction(RD) (or cosine of the angle) between the two vectors ~ a(t);
~
b(t
0
) is given
by
~ a(t) ~
b(t
0
)
j ~ a(t)j j
~
b(t
0
) j
.
9) SR ( ~ a(t);
~
b(t
0
)): Speed Ratio(SR) between the two vectors ~ a(t);
~
b(t
0
) is given by
min j ~ a(t)j; j
~
b(t
0
) j
max j ~ a(t)j; j
~
b(t
0
) j
.
10) R: Transmission range of a mobile node.
11) N: Number of mobile nodes.
12) T : Simulation time.
13) r andom(): returns a value uniformly distributed in the interval [ 1; 1].
B. Mobility Metrics
We propose these metrics to differentiate the various mobility patterns used in our study. The basis of differen-
tiation is the extent to which a given mobility pattern captures the characteristics of spatial dependence, temporal
dependence and geographic restrictions. In addition to these metrics, we also use the Relative Speed metric that
differentiates mobility patterns based on relative motion. This metric was proposed in [13].
1) Degree of Spatial Dependence: It is a measure of the extent of similarity of the velocities of two nodes that
are not too far apart. Formally,
D
spatial
(i; j; t)= RD ( ~ v
i
(t);~ v
j
(t)) SR ( ~ v
i
(t);~ v
j
(t))
The value of D
spatial
(i; j; t) is high when the nodes i and j travel in more or less the same direction and
at almost similar speeds. However, D
spatial
(i; j; t) decreases if the Relative Direction or the Speed Ratio
decreases.
As it is rare for a node’s motion to be spatially dependent on a far off node, we add the condition that
D
i;j
(t) >c
1
R ) D
spatial
(i; j; t)=0
where c
1
> 0 is a constant which will be determined during our experiments in VI.
Average Degree of Spatial Dependence: It is the value of D
spatial
(i; j; t) averaged over node pairs and time
instants satisfying certain condition. Formally,
D
spatial
=
P
T
t=1
P
N
i=1
P
N
j =i+1
D
spatial
(i; j; t)
P
where P is the number of tuples (i; j; t) such that D
spatial
(i; j; t) 6= 0. Thus, if mobile nodes move
independently of one another, then the mobility pattern is expected to have a smaller value for
D
spatial
.
On the other hand, if the node movement is co-ordinated by a central entity, or influenced by nodes in its
neighborhood, such that they move in similar directions and at similar speeds, then the mobility pattern is
expected to have a higher value for
D
spatial
.
April 30, 2003 DRAFT
8
2) Degree of Temporal Dependence: It is a measure of the extent of similarity of the velocities of a node
at two time slots that are not too far apart. It is a function of the acceleration of the mobile node and the
geographic restrictions. Formally,
D
tempor al
(i; t; t
0
)= RD ( ~ v
i
(t);~ v
i
(t
0
)) SR ( ~ v
i
(t);~ v
i
(t
0
))
The value of D
tempor al
(i; t; t
0
) is high when the node travels in more or less the same direction and almost
at the same speed over a certain time interval that can be defined. However, D
tempor al
(i; t; t
0
) decreases if
the Relative Direction or the Speed Ratio decreases.
Arguing in a way similar to that for D
spatial
(i; j; t), we have the following condition
jt t
0
j >c
2
) D
tempor al
(i; t; t
0
)= 0
where c
2
> 0 is a constant which will be determined during our experiments in section VI.
Average Degree of Temporal Dependence: It is the value of D
tempor al
(i; t; t
0
) averaged over nodes and
time instants satisfying certain condition. Formally,
D
tempor al
=
P
N
i=1
P
T
t=1
P
T
t
0
=1
D
tempor al
(i; t; t
0
)
P
where P is the number of tuples (i; t; t
0
) such that D
tempor al
(i; t; t
0
) 6=0 Thus, if the current velocity of
a node is completely independent of its velocity at some previous time step, then the mobility pattern is
expected to have a smaller value for
D
tempor al
. However, if the current velocity is strongly dependent on
the velocity at some previous time step, then the mobility pattern is expected to have a higher value for
D
tempor al
.
3) Relative Speed (RS): We use the standard definition from physics i.e.
RS (i; j; t)= j
~
V
i
(t) ~
V
j
(t) j
As in the case of D
spatial
(i; j; t), we add the following condition
D
i;j
(t) >c
3
R ) RS(i; j; t)= 0
where c
3
> 0 is a constant which will be determined during our experiments in VI.
Average Relative Speed: It is the value of RS (i; j; t) averaged over node pairs and time instants satisfying
certain condition. Formally,
RS =
P
N
i=1
P
N
j =1
P
T
t=1
RS(i; j; t)
P
where P is the number of tuples (i; j; t) such that RS (i; j; t) 6=0.
4) Geographic Restrictions: For this metric, we developed the notion of degree of freedom of points on a map.
Degree of freedom of a point is the number of directions a node can go after reaching that point, but currently
we do not have a good way of quantitatively aggregating this definition for the whole map. Thus, we do not
quantitatively define the Geographic Restrictions, but we qualitatively include it in our study as will be seen
in Section V.
April 30, 2003 DRAFT
9
C. Connectivity Graph Metrics
Since routing protocol performance is in general affected by the network topology dynamics, we feel that it is
useful to have metrics to analyze the effect of mobility on the connectivity graph between the mobile nodes. The
connectivity graph metrics aim to study this effect. These metrics might also help in relating mobility metrics with
protocol performance, which will be shown in Section VI.
The connectivity graph is the graph G =(V; E ), such that jV j = N and at time t, a link (i; j ) 2 E iff D
i;j
(t) R.
Let X (i; j; t) be an indicator random variable which has a value 1 iff there is a link between nodes i and j at time
t. X (i; j)= max
T
t=1
X (i; j; t) be an indicator random variable which is 1 if a link existed between nodes i and j
at any time during the simulation, 0 otherwise.
1) Number of Link Changes: Number of link changes for a pair of nodes i and j is the number of times the
link between them transitions from “down” to “up”. Formally,
LC(i; j)=
T
X
t=1
C (i; j; t)
where C (i; j; t) is an indicator random variable such that C (i; j; t)=1 iff X (i; j; t 1) = 0 and X (i; j; t)= 1
i.e. if the link between nodes i and j is down at time t 1, but comes up at time t.
Average Number of Link Changes: It is the value of LC(i; j ) averaged over node pairs satisfying certain
condition. Formally,
LC =
P
N
i=1
P
N
j =i+1
LC(i; j )
P
where P is the number of pairs i,j such that X (i; j ) 6=0.
2) Link Duration: For two nodes i and j, at time t
1
, duration of the link (i; j ) is the length of the longest time
interval [t
1
,t
2
] during which the two nodes are within the transmission range of each other. Moreover these
two nodes are not within the transmission range at time t
1
and time t
2
+ for > 0. Formally,
LD(i; j; t
1
)= t
2
t
1
iff 8t t
1
t t
2
; > 0 : X (i; j; t) = 1 and X (i; j; t
1
) = 0 and X (i; j; t
2
+ ) = 0. Otherwise,
LD(i; j; t
1
)=0.
Average Link Duration: It is the value of LD(i; j ) averaged over all existing links for node pairs satisfying
certain condition. Formally,
LD =
P
T
t 1 =0
P
N
i=1
P
N
j =i+1
LD (i; j; t
1
)
P
where P is the number of tuples (i; j; t
1
) such that LD(i; j; t
1
) 6=0.
3) Path Duration: For a path P = fn
1
;n
2
; :::n
k
g, consisting of k nodes , at time t
1
, path duration is the length
of the longest time interval [t
1
;t
2
], during which each of the k 1 links between the nodes exist. Moreover, at
time t
1
and time t
2
+ , > 0, at least one of the k links does not exist. Thus, path duration is limited by
the duration of the links along its path. Specifically, at time t
1
, path duration is the minimum of the durations
April 30, 2003 DRAFT
10
of the k 1 links (n
1
;n
2
); (n
2
;n
3
):::(n
k 1
;n
k
) at time t
1
. Formally,
PD (n
1
;n
k
;t
1
)= min
1 z k 1
LD(n
z
;n
z +1
;t
1
)
where PD(n
1
;n
k
;t
1
) is the shortest path between node n
1
and node n
k
at time t
1
.
Average Path Duration: It is the value of PD(n
1
;n
k
;t
1
) averaged over all existing paths for node pairs
satisfying certain condition. Formally,
PD =
P
T
t 1 =0
P
N
n 1 =1
P
N
n
k
=n 1 +1
PD(n
1
;n
k
;t
1
)
P
where P is the number of tuples (n
1
;n
k
;t
1
) such that PD(n
1
;n
k
;t
1
) 6=0.
4) Path Availability: It is the fraction of time during which a path is available between two nodes i and j. The
node pairs of interest are the ones that have communication traffic between them. Formally,
PA(i; j ) =
8
<
:
P
T
t=start(i;j )
A(i;j;t)
T start(i;j )
if T start(i; j ) > 0
0 otherwise
(1)
where A(i; j; t) is an indicator random variable which has a value 1 if a path is available from node i to node
j at time t, and has a value 0 otherwise. start(i; j ) is the time at which the communication traffic between
nodes i and j starts.
Average Path Availability: It is the value of PA(i; j ) averaged over node pairs satisfying certain condition.
Formally,
PA =
P
N
i=1
P
N
j =i+1
PA(i; j )
P
where P is the number of pairs i,j such that T start(i; j ) > 0.
D. Protocol Performance Metrics:
We evaluate the performance of the MANET routing protocols using the metrics of throughput (ratio of the
number of packets delivered to the number of packets sent) and routing overhead (number of routing control
packets sent) as done in several previous studies in this area of research.
V. MOBILITY MODELS
As mentioned in Section I, Random Waypoint does not seem to capture the mobility characteristics of spatial
dependence, temporal dependence and geographic restrictions. In the previous section, we defined Mobility metrics
that either qualitatively or quantitatively define these characteristics. To thoroughly study the effect of mobility on
MANET protocol performance, we seek to evaluate the protocols over a rich set of mobility models that span the
design space of the Mobility metrics. Thus, apart from Random Waypoint, we use the following mobility models:
1) Reference Point Group Mobility (RPGM) Model
2) Freeway Mobility Model
3) Manhattan Mobility Model
April 30, 2003 DRAFT
11
Each of the above models has certain characteristics that are different from Random Waypoint, which will be
shown by our metrics and simulations.
1) RPGM Model: [8] introduced this model. Here, each group has a logical center (group leader) that determines
the group’s motion behavior. Initially, each member of the group is uniformly distributed in the neighborhood
of the group leader. Subsequently, at each instant, every node has a speed and direction that is derived by
randomly deviating from that of the group leader.
Applications: Group mobility can be used in military battlefield communications where the commander and
soldiers form a logical group. More applications are mentioned in [8].
Important Characteristics: Each node deviates its velocity (both speed and direction) randomly from that of
the leader. The movement in group mobility can be characterized as follows:
a)
~
V
member
(t)
=
~
V
leader
(t)
+ r andom() SDR max speed
b) member
(t)= leader
(t)+ r andom() AD R max ang l e
where 0 S D R ; AD R 1. SDR is the Speed Deviation Ratio and ADR is the Angle Deviation Ratio. SDR
and ADR are used to control the deviation of the velocity (magnitude and direction) of group members from
that of the leader. max speed and max angle are used to specify the maximum deviation a group member can
take. In our simulation, we set maximum speed for the group leader as the max speed and set 180
Æ
as the
max angle. Since the group leader mainly decides the mobility of group members, group mobility pattern is
expected to have high spatial dependence for small values of SDR and ADR.
2) Freeway Mobility Model: We propose this new model to emulate the motion behavior of mobile nodes on
a freeway. The freeway map used in our study is shown in Fig.2(a).
Applications: It can be used in exchanging traffic status or tracking a vehicle on a freeway.
Important Characteristics: In this model we use maps. There are several freeways on the map and each freeway
has lanes in both directions. The differences between Random Waypoint and Freeway are the following:
a) Each mobile node is restricted to its lane on the freeway.
b) The velocity of mobile node is temporally dependent on its previous velocity.
c) If two mobile nodes on the same freeway lane are within the Safety Distance (SD), the velocity of the
following node cannot exceed the velocity of preceding node.
The inter-node and intra-node relationships involved are:
a)
~
V
i
(t +1)
=
~
V
i
(t)
+ r andom() j ~ a
i
(t)j
b) 8i; 8j; 8t D
i;j
(t) SD )j
~
V
i
(t)jj
~
V
j
(t)j,if j is ahead of i in its lane.
Due to the above relationships, the Freeway mobility pattern is expected to have spatial dependence and high
temporal dependence. It also imposes strict geographic restrictions on the node movement by not allowing a
node to change its lane.
3) Manhattan Mobility Model: We introduce the Manhattan model to emulate the movement pattern of mobile
nodes on streets defined by maps. The Manhattan map used in our study is shown in Fig.2(b).
April 30, 2003 DRAFT
12
Applications: It can be useful in modeling movement in an urban area where a pervasive computing service
between portable devices is provided.
Important Characteristics: Maps are used in this model too. The map is composed of a number of horizontal
and vertical streets. Each street has two lanes for each direction (North and South direction for vertical streets,
East and West for horizontal streets). The mobile node is allowed to move along the grid of horizontal and
vertical streets on the map. At an intersection of a horizontal and a vertical street, the mobile node can turn
left, right or go straight. This choice is probabilistic: the probability of moving on the same street is 0.5, the
probability of turning left is 0.25 and the probability of turning right is 0.25.
The velocity of a mobile node at a time slot is dependent on its velocity at the previous time slot. Also,
a node’s velocity is restricted by the velocity of the node preceding it on the same lane of the street. The
inter-node and intra-node relationships involved are the same as in the Freeway model.
Thus, the Manhattan mobility model is also expected to have high spatial dependence and high temporal
dependence. It too imposes geographic restrictions on node mobility. However, it differs from the Freeway
model in giving a node some freedom to change its direction.
(a) Freeway Map
(b) Manhattan Map
Fig. 2. Maps used in Freeway and Manhattan Model
Most of the mobility models mentioned above are parameterized. E.g. SDR and ADR are some of the parameters
used in RPGM, while maps are important parameters in the Freeway and Manhattan models. Although we did not
quantitatively define Geographic Restrictions in Section IV, we qualitatively include them in our study by using
the Freeway and Manhattan models. Using a parameterized approach, we aim to get a good coverage of the design
space of the proposed mobility metrics by producing a rich set of mobility patterns that can be used as a “test-suite”
April 30, 2003 DRAFT
13
for further research.
VI. EXPERIMENTS
As a first step, we wanted to validate if our proposed metrics differentiate the mobility models. Once this was
done, we focused on answering the following questions: Whether mobility affects protocol performance?, if yes,
we attempt to answer the questions Why? and How? mentioned in Section I.
A. Validating the Mobility Metrics
Our mobility scenario generator produced the different mobility patterns following the RPGM, Freeway and
Manhattan models according to the format required by ns-2. In all these patterns, 40 mobile nodes moved in an
area of 1000m x 1000m for a period of 900 seconds. Random Waypoint mobility pattern was generated using the
setdest tool which is a part of the ns-2 distribution. For RPGM, we used 2 different mobility scenarios: single group
of 40 nodes and 4 groups of 10 nodes each moving independently of each other and in an overlapping fashion. Both
Speed Deviation Ratio (SDR) and Angle Deviation Ratio (ADR) were set to 0.1. For the Freeway and Manhattan
models, the nodes were placed on the freeway lanes or local streets randomly in both directions initially. Their
movement was controlled as per the specifications of the models. If a node moves beyond the boundary of the area
it is re-inserted at the beginning position in a randomly chosen lane in the area. The maximum speed V
max
was set
to 1, 5, 10, 20, 30, 40, 50 and 60 m/sec to generate different movement patterns for the same mobility model. On
evaluating these patterns with our mobility metrics, we observed that some of the metrics were able to differentiate
between the mobility patterns based on the characteristics we focused on, while the others failed.
Average Relative Speed (
RS): We experimented with different values of the constant c
3
mentioned in Section
IV. For the value of c
3
=2,
RS could differentiate between the different mobility patterns very clearly. As seen in
Fig.3,
RS has the lowest value for RPGM (single group and multiple group mobility) as the nodes move together
in a co-ordinated fashion with little deviation, while it has a medium value for Random Waypoint. Its value for
the Freeway and Manhattan mobility patterns is the highest and almost twice that for Random Waypoint. This high
value is because of the movement in opposite direction for both Freeway and Manhattan mobility patterns.
Average Degree of Spatial Dependence (
D
spatial
): We experimented with different values of the constant c
1
mentioned in Section IV. For the value of c
1
= 2,
D
spatial
could differentiate between the different mobility
patterns very clearly. As seen in Fig.4,
D
spatial
has a higher value for single group mobility (around 0.5) than that
of multiple group mobility (about 0.35). However, for the Random Waypoint, Manhattan and Freeway, its value is
almost 0. Intuitively, in RPGM, the group leader controls the movement of the mobile node and thus the mobility
pattern has a high spatial dependence. Initially, we expected the Freeway and Manhattan mobility patterns to have
a high spatial dependence as a node’s movement is influenced by nodes before it in the lane. Due to the use of
lanes in opposite directions in the map, the positive Degree of Spatial Dependence of a node with nodes in the
same direction cancels the negative Degree of Spatial Dependence of the node with nodes traveling in the opposite
direction.
April 30, 2003 DRAFT
14
0 10203040 50 60
Maximum Speed (m/sec)
0
10
20
30
40
50
Average Relative Speed (m/sec)
Random Waypoint
RPGM (Single Group)
RPGM (4 Groups)
Freeway
Manhattan
Fig. 3. Average Relative Speed
0 10203040 50 60
Maximum Speed (m/sec)
0
0.05
0.1
0.15
0.2
0.25
0.3
0.35
0.4
0.45
0.5
0.55
Average Degree of Spatial Dependence
Random Waypoint
RPGM (Single Group)
RPGM (4 groups)
Freeway
Manhattan
Fig. 4. Average Degree of Spatial Dependence
Average Degree of Temporal Dependence (
D
tempor al
): This metric could not differentiate between the various
mobility patterns used in our study. The usefulness of this metric is still under investigation.
In summary,
RS and
D
spatial
are found to be useful mobility metrics in our study. Fig.3 and Fig.4 show that
for each of these metrics, we had scenarios with relatively low values, medium values and relatively high values.
Similarly, for Geographic Restrictions, the Freeway does not allow a node to change directions as freely as the
Manhattan model. So, we believe that our “test-suite” has given a reasonably good coverage of the mobility metric
space.
B. Validating the Connectivity Graph Metrics
To study the effect of mobility on the Connectivity Graph, we evaluated the connectivity graphs resulting from the
mobility patterns used in Section VI-A. We had the following observations about the Connectivity Graph metrics:
Average Link Duration (
LD): As seen in Fig.5,
LD has a higher value for single group and multiple groups than
Random Waypoint. For the Freeway and Manhattan its value is similar to Random Waypoint or even worse. Since
nodes in a group move at velocities that are deviated by a small fraction from the group leader, an already existing
link between two nodes is expected to have a higher duration. The low value for the Freeway and Manhattan may
April 30, 2003 DRAFT
15
0 10203040 50 60
Maximum Speed (m/sec)
0
200
400
600
800
1000
Average Link Duration (sec)
Random Waypoint
RPGM (Single Group)
RPGM (Multiple Groups)
Freeway
Manhattan
Fig. 5. Average Link Duration
0 10203040 50 60
Maximum Speed (m/sec)
0
200
400
600
800
1000
Average Path Duration (sec)
Random Waypoint
RPGM(single group)
RPGM(multiple groups)
Freeway
Manhattan
Fig. 6. Average Path Duration
be because of the opposite direction of motion and high relative speeds.
Average Path Duration (
PD): The path duration is counted as the interval between the time when the path is
set up and the time when the path is broken. However, there can be potentially exponential paths between any
specific source - destination pair. Analyzing the duration of all these paths might not be feasible. As a reasonable
approximation, we define the path duration as the duration of the shortest path. The shortest path between the
source and the destination is computed by the Breadth First Search(BFS) algorithm [19]. Similar to
LD, as shown
in Fig.6, RPGM(single group and multiple groups) has a higher
PD value than Random Waypoint. The
PD values
for Manhattan and Freeway are similar to or slightly lower than Random Waypoint. Since each path is composed
of several links, it is likely that the behavior of path duration will be determined by the behavior of link duration.
Average Number of Link Changes(
LC): This metric was not able to differentiate between the several mobility
patterns used in our study.
Average Path Availability(
PA): We use the Breadth First Search algorithm on the snapshots of the network to
calculate whether a path between a specific source destination pair exists [19]. For RPGM (single group), RPGM
(mutiple group), Random Waypoint, Freeway and Manhattan models,
PA is found to be around 100%, 92%, 97%,
99% and 95% respectively. In most cases, a path is available at least 95% of the time. Thus, the difference across
the models was too small to be of any help.
April 30, 2003 DRAFT
16
In summary,
LD and
PD are found to be useful metrics to differentiate the connectivity graph arising from the
different mobility patterns used in our study.
C. Whether mobility affects protocol performance?
To evaluate the effect of mobility on the performance of protocols, we carried out simulations in the network
simulator (ns-2) environment with the CMU Wireless Ad Hoc networking extension. The transmission range of
the nodes was 250m. The mobility patterns used were the same as those used to Section VI-A. The traffic pattern
was generated by the cbrgen tool that is part of ns-2 distribution. The traffic consisted of 20 Constant Bit Rate
(CBR) sources and 30 connections. The source-destination pairs were chosen at random. The data rate used was 4
packets/sec and the packet size was 64 bytes.
To remove any effects due to randomness of the traffic pattern, we used different random seeds to generate 3
different traffic patterns having the same number of sources and connections. The results for each model (for a
given V
max
) are averaged over simulation runs using these 3 different traffic patterns.
We evaluated the performance of DSR, AODV and DSDV across this rich set of mobility models and observed
that the mobility models may drastically affect protocol performance. We use DSR as an illustrative example.
0 10203040 50 60
Maximum Speed (m/sec)
50
60
70
80
90
100
Throughput (%)
Random Waypoint
RPGM (Single Group)
RPGM (4 Groups)
Freeway
Manhattan
(a) Throughput
0 10203040 50 60
Maximum Speed (m/sec)
0
20000
40000
60000
80000
Routing Overhead (packets)
Random Waypoint
RPGM (Single Group)
RPGM (4 Groups)
Freeway
Manhattan
(b) Routing Overhead
Fig. 7. DSR Protocol Performance across Mobility Models
DSR shows a difference of almost 40% in throughput from Manhattan to the RPGM (Single Group) model as
April 30, 2003 DRAFT
17
seen from Fig.7(a).
Also, there is an order of magnitude difference in the routing overhead of DSR across the various models as shown
by Fig.7(b). Similar performance differences were observed for other protocols used in our study. We observed that
DSR, DSDV and AODV achieve the highest throughput and the least overhead with RPGM and incur high overhead
and low throughput with both Freeway and Manhattan models. This is consistent with the observations made in
[8] which evaluated the protocols using Random Waypoint and several other group mobility applications. However,
we take a step further and attempt to analyze the reason for this performance difference in Section VI-D.
Relative Performance of Protocols Across Mobility Models: In this part, we investigated the effect of mobility on
relative rankings of protocol performance. As shown in Fig.8(a), Fig.9(a), Fig.10(a), Fig.11(a) and Fig.12(a), DSR
seems to produce the highest throughput in most cases, while AODV seems to outperform DSR (by almost 31%)
in the Manhattan model. As seen from Fig.9(a) and Fig.12(a), the relative ranking of AODV and DSDV in terms
of throughput seems to depend on the underlying mobility model.
0 10203040 50 60
Maximum Speed (m/sec)
50
60
70
80
90
100
Throughput (%)
DSDV
AODV
DSR
(a) Random Waypoint: Throughput
0 10203040 50 60
Maximum Speed (m/sec)
0
25000
50000
75000
1e+05
1.25e+05
1.5e+05
Routing Overhead (packets)
DSDV
AODV
DSR
(b) Random Waypoint: Routing Overhead
Fig. 8. Protocol Performance for Random Waypoint Model
Also, DSR incurs the least routing overhead in most cases, while DSDV has a lower overhead than DSR in
the Freeway and Manhattan models as shown in Fig.11(b) and Fig.12(b). The relative ranking of DSR and DSDV
in terms of routing overhead seems to depend on the underlying mobility model as shown in Fig.8(b), Fig.9(b),
Fig.10(b), Fig.11(b) and Fig.12(b).
April 30, 2003 DRAFT
18
0 10203040 50 60
Maximum Speed (m/sec)
50
60
70
80
90
100
Throughput (%)
DSDV
AODV
DSR
(a) RPGM(single group): Throughput
0 10203040 50 60
Maximum Speed (m/sec)
0
2000
4000
6000
8000
10000
Routing Overhead (packets)
DSDV
AODV
DSR
(b) RPGM(single group): Routing Overhead
Fig. 9. Protocol Performance for RPGM(single group) Model
Thus, we conclude that relative rankings of protocols may vary with the mobility model used. We also observe
that DSDV achieves a higher throughput than AODV (by around 10%) in RPGM. Thus, in general it is not always
true that on demand protocols perform better than table driven ones in terms of throughput. Also, a protocol with
the least overhead does not always produce the highest throughput. E.g. in the Freeway model, DSDV seems to
have the least throughput and the least overhead.
Although, these results were somewhat expected, the quantitative analysis helped us gain a lot of insight to
answer the next question.
D. Why mobility affects protocol performance?
First, the relationship between the mobility metrics and the performance metrics was unclear. But after introducing
the connectivity graph metrics, we were able to observe a very clear correlation between Average Degree of Spatial
Dependence, Average Relative Speed, Average Link Duration, Average Path Duration and protocol performance
metrics. The mobility pattern influences the connectivity graph which in turn influences the protocol performance.
The relationship is identified in Fig.1.
In general, it was observed that DSR, DSDV and AODV had a higher throughput and lower overhead for the
group mobility models than for the Random Waypoint model. At the same time, all the protocols had a higher
April 30, 2003 DRAFT
19
0 10203040 50 60
Maximum Speed (m/sec)
50
60
70
80
90
100
Throughput (%)
DSDV
AODV
DSR
(a) RPGM(4 groups): Throughput
0 10203040 50 60
Maximum Speed (m/sec)
0
2000
4000
6000
8000
10000
Routing Overhead (packets)
DSDV
AODV
DSR
(b) RPGM(4 groups): Routing Overhead
Fig. 10. Protocol Performance for RPGM(4 groups) Model
throughput and lower overhead for Random Waypoint than the Freeway and Manhattan models. One plausible
reason for this observation can be as follows:
1) With similar relative speed, between Random Waypoint and RPGM, high degree of spatial dependence (for
RPGM) means higher link duration and correspondingly higher path duration, which in turn will result in
higher throughput and lower routing overhead.
2) With the same degree of spatial dependency, between Freeway/Manhattan and Random Waypoint, high relative
speed (for Freeway/Manhattan) means lower link duration and correspondingly lower path duration, which
will result in lower throughput and higher overhead.
The above reasoning can be explained as follows: For a given relative speed, if a mobility pattern has a high
degree of spatial dependence, an already existing link between two nodes is expected to remain stable for a longer
period of time as the nodes are likely to move together, then the path between source and destination is relatively
stable. Thus fewer packets will be dropped due to path breakage leading to higher throughput. At the same time, the
control overhead is lower as little effort is needed to repair the seldom broken path. For a given spatial dependence,
if a mobility pattern has a high relative speed, the nodes might move out of range more quickly. Thus an already
existing link and the paths using this specific link may remain stable for a relatively shorter duration. This may
lead to more packets being dropped due to path breakage, resulting in lower throughput. Higher control overhead is
April 30, 2003 DRAFT
20
0 10203040 50 60
Maximum Speed (m/sec)
50
60
70
80
90
100
Throughput (%)
DSDV
AODV
DSR
(a) Freeway: Throughput
0 10203040 50 60
Maximum Speed (m/sec)
0
25000
50000
75000
1e+05
1.25e+05
1.5e+05
Routing Overhead (packets)
DSDV
AODV
DSR
(b) Freeway: Routing Overhead
Fig. 11. Protocol Performance for Freeway Model
needed to repair the more frequently broken path. We also note that the Freeway and Manhattan mobility patterns
have high relative speed and low degree of spatial dependence leading to the worst performance of all the protocols
while using these models.
VII. ANALYSIS OF BUILDING BLOCKS
Unlike the conventional evaluation studies, we pursue our analysis beyond the “whole protocol” level and attempt
to answer How mobility affects protocol performance by looking into the “parts” that constitute the MANET routing
protocols. We propose an approach to systematically decompose a protocol into its functional mechanistic ”building
blocks”. Each building block can be thought of as a parameterized ”black box”. The parameter settings define the
behavior of each block, while the behavior of building blocks and the nature of interaction between the building
blocks defines the behavior of the protocol as a ”whole”. We use the analysis of reactive protocols as an example
to illustrate this approach. In this section, we carry out a preliminary analysis of the impact of mobility on two
reactive routing protocols after identifying the basic building blocks of MANET reactive routing protocols and
their parameter setting. Thus we can extract the relative merits of different parameter settings and achieve a better
understanding of various building blocks of MANET routing protocols, which will serve as a solid cornerstone for
development of more efficient MANET routing protocols.
April 30, 2003 DRAFT
21
0 10203040 50 60
Maximum Speed (m/sec)
50
60
70
80
90
100
Throughput (%)
DSDV
AODV
DSR
(a) Manhattan: Throughput
0 10203040 50 60
Maximum Speed (m/sec)
0
25000
50000
75000
1e+05
1.25e+05
1.5e+05
Routing Overhead (packets)
DSDV
AODV
DSR
(b) Manhattan: Routing Overhead
Fig. 12. Protocol Performance for Manhattan Model
The part(a) and part(b) of Figure 13 show the building block architecture for DSR and AODV respectively, the
part(c) of Figure 13 shows a generalized building block architecture for reactive MANET protocols.
A. Design Choices of Building Blocks for DSR and AODV
First we discuss the design choices (parameter settings) of the identified building blocks of reactive MANET
routing protocols and specific parameter settings for DSR and AODV . We pose some questions about the utility of
the various design choices made by these protocols. In section VII-B, we attempt to answer these questions.
The mechanism of reactive MANET routing protocols such as DSR and AODV
1
is composed of two major
phases:
1) Route Setup Phase:: Route Discovery is the major mechanism in this phase. It is initiated if there is no
cached route available to the destination. This mechanism consists of the following building blocks:
1) Flooding building block: The flooding building block takes responsibility to distribute the route request
messages within the network. Here, the key parameter is the range of flooding, generally described by TTL
1
Current AODV implementation in ns 2(version ns-2.1b8a) and IETF specification adopt the expanding ring search for query flooding,
localized rediscovery for error handling and source-specific error notification, which may be different with the original AODV paper. In our
study, if not particularly specified, we refer the implementation in ns 2(version ns-2.1b8a) as the standard AODV mechanism.
April 30, 2003 DRAFT
22
DSR AODV
Local Inquiry &
Global Flooding
Cache
Management
Link
Monitoring
Salvaging
Error
Notification
Expanding Ring
Search & Global
Flooding
Cache
Management
Link
Monitoring
Localized
Rediscovery
Error
Broadcast
(a) (b)
Route Setup
Route Maintenance
Flooding Caching
Range of Flooding
Caching Style
Expiration Timer
Error
Detection
Error
Handling
Error
Notification
Detection
Method
Handling
Mode
Recipient
Route Request
Add Route Cache
Route Reply
Link Breaks Notify
Route
Invalidate
Localized/Non-localized method
Notify
Generalization of
Error Handling
(c)
Generalization of
Error Handling
Generalization
of Flooding
Generalization
of Flooding
Fig. 13. Diagram of Building Block Framework for Reactive Protocols
field in the IP header. If the TTL is set to the network diameter, a global flooding is done. One optimization
is to implement the localized controlled flooding before global flooding. This may be useful if the probability
of finding an appropriate cache in the neighborhood is high.
For the range of flooding, both DSR and AODV use the two-step controlled flooding. DSR conducts a non-
propagating direct-neighborhood inquiry(TTL=1) before the global flooding(TTL=D, D is network diameter).
Similarly, AODV uses the expanding ring search(TTL=1,3,5,7) before the global flooding is initiated. Here,
we want to answer the following question: How useful are non-propogating route requests?
2) Caching building block: The caching building block helps to efficiently and promptly provide the route
to the destination without referring to the destination every time. Several parameters affect the behavior of
the caching building block. One of those parameters is whether aggressive caching is allowed, i.e. whether
multiple cache entries are allowed for the same destination and whether a node can cache the route information
it overhears? Generally speaking, aggressive caching scheme increases the possibility of finding an appropriate
route without re-initiating a route discovery.
DSR uses aggressive caching, while AODV does not. For caching, We are interested in the following questions:
How useful is caching? and Is aggressive caching better than non-aggressive caching?
2) Route Maintenance Phase:: Route Maintenance phase takes the responsibility of detecting broken links and
repairing the corresponding routes. This phase is made up of the following building blocks:
1) Error Detection building block: It is used to monitor the status of the link of a node with its immediate neigh-
bors. Several methods to monitor the link status between neighbors can be used: MAC level acknowledgement,
network-layer explicit Hello message or network-layer passive overhearing scheme. Here, the parameter is
the mode of error detection used.
For mode of error detection, both DSR and AODV can use either of the three choices. Since all these schemes
April 30, 2003 DRAFT
23
are very similar, we do not investigate this building block in our analysis.
2) Error Handling building block: It finds alternative routes to replace an invalid route after a broken link is
detected. One of the parameters to this block is what recovery scheme should be used. In a localized recovery,
the node detecting the broken link will attempt to find an alternative route in its own cache or do a localized
flooding before asking the source to re-initiate the route discovery.
For localized recovery, in DSR, on detecting a broken link, the upstream node will first search its cache
to replace the invalid route(this scheme is called salvaging), although the found alternative route may also
be invalid in some scenarios. While in AODV , the upstream node detecting the broken link will initiate a
localized flooding to find the route to the destination. For this building block, we are interested in the following
question: Which is a better scheme for localized error handling: cache lookup or localized flooding?
3) Error Notification building block: It is used to notify the nodes in the network about invalid routes. The
key parameter to this building block is the recipient of the error message. Either only the source is notified
or the entire network is notified.
For recepient of error notification, both DSR and AODV notify the error to the source. Since both DSR and
AODV use the same parameter setting, we do not investigate this block during our simulations.
Besides these three questions about the design choices, we are also interested at the plausible explanation for the
observation we made in section VI: DSR outperforms AODV in most mobility scenarios except the Freeway and
Manhattan model with high mobility.
B. Performance Evaluation
We identified parts of the network simulator (ns-2) code [1] which implement these building blocks and profiled
them during our simulations. The simulation setting is exactly the same as section VI.
1) Flooding: Fig.14(a) and Fig.14(b) show the ratio of non propagating route requests to the total number of
route requests issued by the DSR and AODV respectively. This metric measures the likelihood of finding a route to
the destination from the source’s neighborhood. The result shows that non-propagating route request is frequently
used (more than 30% for DSR and more than 10% for AODV in most scenarios).
It is observed that this ratio increases from Random Waypoint to Manhattan to Freeway mobility models. This is
because the geographic constraints on movement are greater in Freeway than Manhattan, which are in turn greater
than in Random Waypoint. Thus, the likelihood of finding a route to the destination from the source’s neighbors
increases from Random Waypoint to Manhattan to Freeway.
On the other hand, the ratio for DSR is almost twice as large as that for AODV across all mobility models. A
possible reason for this comes from the fact that DSR uses aggressive caching as compared to AODV . When such
a caching scheme is coupled with the mechanism of non propagating route requests, it translates to low routing
overhead and high throughput as was shown in our study and several other comparative studies.
Thus, it seems that caching has a significant impact on the performance of DSR and AODV . Hence we study it
next.
April 30, 2003 DRAFT
24
0 10203040 50 60
Maximum Speed (m/sec)
0
0.2
0.4
0.6
0.8
# of Non-Propogating Route Reqs. / Total # of Route Reqs.
Random Waypoint
Manhattan
Freeway
RPGM (Single Group)
(a) DSR
0 10203040 50 60
0
0.1
0.2
0.3
0.4
Random Waypoint
Manhattan
Freeway
RPGM (1 Group)
RPGM (4 Groups)
(b) AODV
Fig. 14. Ratio of Non Propogating Route Requests to Total Number of Route Requests
2) Caching: To measure the effectiveness of caching, we evaluate the ratio of the number of route replies coming
from the cache to the total number of route replies.
Fig. 15(a) and Fig.15(b) show that this ratio is high for Random Waypoint, Manhattan and Freeway models,
which implies that most of the route replies for these mobility models come from the cache(more than 80% in most
mobility scenarios).
The difference in the ratio for DSR and AODV is greater than 20% for all mobility models. DSR uses aggressive
caching as compared to AODV . Thus, the likelihood of a route reply coming from a cache is higher in DSR than
in AODV . Thus, fewer route requests will be needed and thus the routing overhead of DSR is lower than AODV
as we observed in Section VI. Thus, aggressive caching seems to be a good design choice.
To completely evaluate the caching strategy, we also need to examine the validity of the cache entries. We evaluate
the ratio of invalid cache entries to the total number of cache entries for DSR.
As shown by Fig. 16, the ratio increases from RPGM (around 10%) to Random Waypoint to Freeway (around
60%) to Manhattan (around 80%) mobility models. Thus caching may have adverse effects in mobility models with
a high relative speed which may lead to cache invalidation. Packets may be sent on invalid routes which might lead
April 30, 2003 DRAFT
25
0 10203040 50 60
Maximum Speed (m/sec)
0
0.2
0.4
0.6
0.8
1
# Route Replies from Cache / Total # of Route Replies
Random Waypoint
Manhattan
Freeway
RPGM (Single Group)
RPGM (4 Groups)
(a) DSR
0 10203040 50 60
Maximum Speed (m/sec)
0.2
0.4
0.6
0.8
1
# of Route Replies from the Cache / Total # of Route Replies
Random Waypoint
Manhattan
Freeway
RPGM (Single Group)
RPGM (4 Groups)
(b) AODV
Fig. 15. Ratio of the Number of Route Replies from the Cache to the Total Number of Route Replies
to packets being dropped and route request retries. This leads to a lower throughput and higher overhead for DSR
for the Freeway and Manhattan models as was shown in our study.
On the other hand, in mobility models with very high relative speed like Manhattan and Freeway, AODV seems
to achieve as good a throughput as DSR (and sometimes better). AODV does not use aggressive caching, thus the
ratio of the number of route replies coming from the cache to the total number of route replies is lesser for AODV
than DSR. Thus, the likelihood of getting invalid routes from the cache is lesser for AODV than for DSR. This
may explain why AODV outperforms DSR in Freeway and Manhattan models with high mobility.
Moreover, at high relative speeds, the number of routes broken is greater. Thus, a protocol which has a better
error handling mechanism at higher relative speeds might perform better in such situations. This line of reasoning
leads us to evaluate the next building block of interest - Error Handling.
3) Error Handling: To study the effectiveness of error handling, we focus on localized error handling. We
evaluate the ratio of the number of localized error handling to the total number of route errors for both DSR and
AODV . For DSR, we notice that salvaging accounts for less than 2% of the total number of route errors. Moreover,
if we take invalid cache entries into account, the effect of salvaging on the protocol performance is further lowered.
On the other hand, in AODV , a route request is initiated by the upstream node which detects the broken link if it is
April 30, 2003 DRAFT
26
0 10203040 50 60
Maximum Speed (m/sec)
0
0.2
0.4
0.6
0.8
1
# of Invalid Cache Replies / Total # of Cache Replies
Random Waypoint
Manhattan
Freeway
RPGM (Single Group)
RPGM (Multiple Groups)
Fig. 16. Ratio of the Number of Invalid Route Replies from the Cache to the Total Number of Route Replies from the Cache
0 10203040 50 60
Maximum Speed (m/sec)
0
0.1
0.2
0.3
0.4
0.5
# of Localized Route Requests / Total # of Route Errors
Random Waypoint
Manhattan
Freeway
RPGM (Single Group)
RPGM (4 Groups)
Fig. 17. Ratio of the Number of Localized Route Recovery Requests to the Total Number of Route Errors for AODV
closer to the destination. As Fig.17 shows, the ratio is between 40% and 50% for Freeway and Manhattan models.
Moreover the routes obtained by this mechanism are more up to date than those from the cache salvaging in DSR.
This is another factor which explains the better performance of AODV as compared to DSR in the Freeway and
Manhattan models.
C. Discussion
The above study of the building blocks has given us greater insight into the design of the reactive routing protocols
for MANETs. Decomposing a protocol into building blocks and evaluating these building blocks have shown us
the scenarios in which the chosen parameters can give a better performance. From the above study, we learnt the
following principles of protocol design:
1) Caching helps reduce the protocol overhead. However, whether aggressive caching should be used depends
on the scenarios in which the protocol will be deployed. For low mobility scenarios, aggressive caching might
be useful, while for higher mobility scenarios, the more stale cache entries incurred by aggressive caching
might affect the protocol throughput.
2) Non Propogating route requests, when combined with caching also reduce the protocol overhead. If caching is
widely done in the network, it may be more advantageous to do non propagating route requests (or expanding
ring search) than globally flooding the route request. In DSR, due to aggressive caching, it may be more
April 30, 2003 DRAFT
27
useful to do expanding ring search (from the source) on a route error than doing a global flooding (from the
source). Again this might work well only for low mobility scenarios.
3) The nature of localized error handling also has a significant impact on protocol performance. Re-initiating a
route request from an intermediate node can be more advantageous than doing a local cache lookup in high
mobility scenarios, while a cache lookup might be more advantageous for low mobility scenarios.
Thus, no particular parameter setting of these building blocks is the most optimal for all scenarios. This further
strengthens our conclusion that there is no clear winner among the protocols across all mobility scenarios.
VIII. CONCLUSIONS &FUTURE WORK
In this paper, we proposed a framework to analyze the impact of mobility pattern on routing performance of
mobile ad hoc network in a systematic manner. In our study, we observe that the mobility pattern does influence the
performance of MANET routing protocols. This conclusion is consistent with the observation of previous studies.
But unlike previous studies that compared different ad hoc routing protocols, there is no clear winner among the
protocols in our case, since different mobility patterns seem to give different performance rankings of the protocols.
We hope that our “test-suite” of mobility models can be incorporated into the current scenarios used to test the
MANET routing protocols.
Moreover, we observe that the mobility pattern influences the connectivity graph that in turn influences the
protocol performance. In addition, we did a preliminary investigation of the common building blocks of MANET
routing protocols, the effect of mobility on these building blocks and how they influence the protocol as a “whole”.
In the future, we plan to study the impact of our “test-suite” on the performance of other ad hoc network
protocols like multicast ad hoc, geographic routing protocols. This study would help us understand the impact of
mobility more deeply and clearly. We believe that several parameters such as traffic patterns, node density and
initial placement pattern of nodes may affect the routing performance and need to investigate them further.
REFERENCES
[1] L. Breslau, D. Estrin, K. Fall, S. Floyd, J. Heidemann, A. Helmy, P. Huang, S. McCanne, K. Varadhan, Y . Xu, and H. Yu, “Advances in
network simulation,” in IEEE Computer, vol. 33, no. 5, May 2000, pp. 59–67.
[2] F. Bai, N. Sadagopan, and A. Helmy, “Important: a framework to systematically analyze the impact of mobility on performance of routing
protocols for adhoc networks,” in INFOCOM 2003, Apr. 2003.
[3] N. Sadagopan, F. Bai, B. Krishnamachari, and A. Helmy, “Paths: analysis of path duration statistics and their impact on reactive manet
routing protocols,” in MobiHoc 2003, Jun. 2003.
[4] F. Bai, N. Sadagopan, and A. Helmy, “Brics: A building-block approach for analyzing routing protocols in ad hoc networks – a case study
of reactive routing protocols,” in USC-CS-TR-02-775, Nov. 2002.
[5] D. B. Johnson, D. A. Maltz, and J. Broch, “DSR: The dynamic source routing protocol for multi-hop wireless ad hoc networks,” in Ad
Hoc Networking, C. Perkins, Ed. Addison-Wesley, 2001, pp. 139–172.
[6] C. E. Perkins and P. Bhagwat, “Highly dynamic destination sequenced distance vector routing (DSDV) for mobile computers,” in ACM
SIGCOMM, 1994, pp. 234–244.
[7] C. Perkins, “Ad hoc on demand distance vector (AODV) routing, internet draft, draft-ietf-manet-aodv-00.txt.”
[8] X. Hong, M. Gerla, G. Pei, and C.-C. Chiang, “A group mobility model for ad hoc wireless networks,” in ACM/IEEE MSWiM, August
1999.
April 30, 2003 DRAFT
28
[9] J. Broch, D. A. Maltz, D. B. Johnson, Y .-C. Hu, and J. Jetcheva, “A performance comparison of multi-hop wireless ad hoc network routing
protocols,” in Proceedings of the Fourth Annual ACM/IEEE International Conference on Mobile Computing and Networking, ACM, October
1998.
[10] S. R. Das, C. E. Perkins, and E. M. Royer, “Performance comparison of two on-demand routing protocols for ad hoc networks,” in
INFOCOM, March 2000.
[11] S. R. Das, R. Castaneda, J. Yan, and R. Sengupta, “Comparative performance evaluation of routing protocols for mobile, ad hoc networks,”
October 1998, pp. 153–161.
[12] V . D. Park and M. S. Corson, “Temporally-ordered routing algorithm (TORA) version 1: Functional specification, internet-draft, draft-ietf-
manet-tora-spec-01.txt,” August 1998.
[13] P. Johansson, T. Larsson, N. Hedman, B. Mielczarek, and M. Degermark, “Scenario-based performance analysis of routing protocols for
mobile ad-hoc networks,” in International Conference on Mobile Computing and Networking (MobiCom’99), 1999, pp. 195–206.
[14] G.Pei, M. Gerla, X. Hong, and C.-C.-Chiang, “A wireless hierarchical protocol with group mobility,” in IEEE WCNC, September 1999.
[15] X. Hong, T. Kwon, M. Gerla, D. Gu, and G. Pei, “A mobility framework for ad hoc wireless networks,” in ACM Second International
Conference on Mobile Data Management (MDM), January 2001.
[16] D. A. Maltz, J. Broch, J. Jetcheva, and D. B. Johnson, “The effects of on-demand behavior in routing protocols for multi-hop wireless ad
hoc networks,” in IEEE Journal on Selected Areas in Communications special issue on mobile and wireless networks, August 1999.
[17] W. Su, S.-J. Lee, and M. Gerla, “Mobility prediction in wireless networks,” in IEEE MILCOM, October 2000.
[18] S.-J. Lee, M. Gerla, and C.-K. Toh, “A simulation study of table-driven and on-demand routing protocols for mobile ad hoc networks,”
vol. 13, no. 4, August 1999, pp. 48–54.
[19] T. H. Cormen, C. E. Leiserson, R. L. Rivest, and C. Stein, “Introduction to algorithms.” McGraw Hill, 1998.
April 30, 2003 DRAFT

Linked assets

Computer Science Technical Report Archive

Conceptually similar

PDF

USC Computer Science Technical Reports, no. 765 (2002)

PDF

USC Computer Science Technical Reports, no. 775 (2002)

PDF

USC Computer Science Technical Reports, no. 778 (2002)

PDF

USC Computer Science Technical Reports, no. 749 (2001)

PDF

USC Computer Science Technical Reports, no. 860 (2005)

PDF

USC Computer Science Technical Reports, no. 803 (2003)

PDF

USC Computer Science Technical Reports, no. 809 (2003)

PDF

USC Computer Science Technical Reports, no. 806 (2003)

PDF

USC Computer Science Technical Reports, no. 856 (2005)

PDF

USC Computer Science Technical Reports, no. 797 (2003)

PDF

USC Computer Science Technical Reports, no. 804 (2003)

PDF

USC Computer Science Technical Reports, no. 801 (2003)

PDF

USC Computer Science Technical Reports, no. 837 (2004)

PDF

USC Computer Science Technical Reports, no. 696 (1999)

PDF

USC Computer Science Technical Reports, no. 858 (2005)

PDF

USC Computer Science Technical Reports, no. 781 (2002)

PDF

USC Computer Science Technical Reports, no. 657 (1997)

PDF

USC Computer Science Technical Reports, no. 770 (2002)

PDF

USC Computer Science Technical Reports, no. 816 (2004)

PDF

USC Computer Science Technical Reports, no. 790 (2003)

Description

Fan Bai, Narayanan Sadagopan, Ahmed Helmy. "The IMPORTANT framework for analyzing the impact of mobility on performance of routing protocols for adhoc networks." Computer Science Technical Reports (Los Angeles, California, USA: University of Southern California. Department of Computer Science) no. 789 (2003).

Asset Metadata

Creator Bai, Fan (author), Helmy, Ahmed (author), Sadagopan, Narayanan (author)

Core Title USC Computer Science Technical Reports, no. 789 (2003)

Alternative Title The IMPORTANT framework for analyzing the impact of mobility on performance of routing protocols for adhoc networks (title)

Publisher Department of Computer Science,USC Viterbi School of Engineering, University of Southern California, 3650 McClintock Avenue, Los Angeles, California, 90089, USA (publisher)

Tag OAI-PMH Harvest

Format 28 pages (extent), technical reports (aat)

Language English

Unique identifier UC16270400

Identifier 03-789 The IMPORTANT framework for analyzing the Impact of Mobility on Performance of RouTing protocols for Adhoc NeTworks (filename)

Legacy Identifier usc-cstr-03-789

Format 28 pages (extent),technical reports (aat)

Rights Department of Computer Science (University of Southern California) and the author(s).

Internet Media Type application/pdf

Source 20180426-rozan-cstechreports-shoaf (batch), Computer Science Technical Report Archive (collection), University of Southern California. Department of Computer Science. Technical Reports (series)

Access Conditions The author(s) retain rights to their work according to U.S. copyright law. Electronic access is being provided by the USC Libraries, but does not grant the reader permission to use the work if the desired use is covered by copyright. It is the author, as rights holder, who must provide use permission if such use is covered by copyright.

Repository Name USC Viterbi School of Engineering Department of Computer Science

Repository Location Department of Computer Science. USC Viterbi School of Engineering. Los Angeles\, CA\, 90089

Repository Email csdept@usc.edu

Inherited Values

Title Computer Science Technical Report Archive

Description Archive of computer science technical reports published by the USC Department of Computer Science from 1991 - 2017.

Coverage Temporal 1991/2017

Repository Email csdept@usc.edu

Repository Name USC Viterbi School of Engineering Department of Computer Science

Repository Location Department of Computer Science. USC Viterbi School of Engineering. Los Angeles\, CA\, 90089

Publisher Department of Computer Science,USC Viterbi School of Engineering, University of Southern California, 3650 McClintock Avenue, Los Angeles, California, 90089, USA (publisher)