university of southern california paths: analysis of path duration statistics and their impact on...
TRANSCRIPT
UNIVERSITY OFSOUTHERN CALIFORNIA
PATHS: Analysis of PATH Duration Statistics and their Impact on Reactive
MANET Routing Protocols
Narayanan Sadagopan*, Fan Bai+,
Bhaskar Krishnamachari*+, Ahmed Helmy+
* Computer Science Department
+ Electrical Engineering Department
University of Southern California
UNIVERSITY OFSOUTHERN CALIFORNIA
Outline• Introduction and Motivation• Background on Mobility Models• Link/Path Duration Metrics• Results & Observations
– Distributions at low and high mobility
– Is it exponential?!
• Models: Path Duration, Throughput & Overhead• Conclusions and Future Work
UNIVERSITY OFSOUTHERN CALIFORNIA
Introduction and Motivation
• Mobility affects performance of MANET protocols significantly [IMPORTANT ‘Infocom BSH03’]
• Mobility affects connectivity, and in turn protocol mechanisms and performance
• In this study: – Closer look at the mobility effects on connectivity metrics
(statistics of link duration (LD) and path duration (PD))
– Develop approximate expressions for LD & PD distributions (Is it really exponential? When is it exponential?)
– Develop first order models for Tput & Overhead as f(PD)
Mobility Connectivity
Protocol MechanismsPerformance(Throughput,
Overhead)
UNIVERSITY OFSOUTHERN CALIFORNIA
Mobility Models
• Used the IMPORTANT framework and tools[BSH’03]• Rich set of mobility models that exhibit various spatial
correlation and relative velocities• Four main models:
– Random Waypoint (RW) [CMU Monarch, BMJHJ’98]
– Reference Point Group Mobility (RPGM) [UCLA, HGPC’99]
– Freeway (FW)
– Manhattan (MH)
UNIVERSITY OFSOUTHERN CALIFORNIA
max
max())()(
()|)(||)(|
ADRrandomttVSDRrandomtVtV
referencenode
referencenode
Mobility Models (contd.)• Random Waypoint Model (RWP)
– A node picks random destination & random velocity [0, Vmax]
– After reaching the destination, it stops for the “pause time”.
– This procedure is repeated until simulation ends
• Reference Point Group Mobility (RPGM)– A group’s general movement is determined by a logical reference point
– Each node in a group follows the reference point with small deviation:
– Angle Deviation Ratio(ADR) and Speed Deviation Ratio(SDR). • In our study ADR=SDR=0.1
– Two scenarios: Single Group (SG) and Multiple Group (MG)
UNIVERSITY OFSOUTHERN CALIFORNIA
Map for FW
Map for MH
Mobility Models (contd.)• Freeway Model (FW)
– A node is restricted to its lane on the freeway
– Velocity of a node is temporally dependent on its previous velocity
– 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
• Manhattan Model (MH)– Similar to Freeway model
– Allows nodes to make turns at intersections
UNIVERSITY OFSOUTHERN CALIFORNIA
Connectivity Metrics
• Link Duration (LD): – For nodes i,j, the duration of link i-j is the longest interval
in which i & j are directly connected
– LD(i,j,t1)=t2-t1
• iff t, t1 t t2, > 0 : X(i,j,t)=1,X(i,j,t1-)=0, X(i,j,t2+)=0
• Path Duration (PD):– Duration of path P={n1,n2,…,nk} is the longest interval in
which all k-1 links exist
UNIVERSITY OFSOUTHERN CALIFORNIA
Simulation Scenarios in NS-2• Path duration computed for the shortest path (at graph
level) until it breaks.Validated later via protocol paths.• Mobility Models (IMPORTANT tool)
– Vmax = 1,5,10,20,30,40,50,60 m/s,
– RPGM: 4 groups (called RPGM4)
– Speed/Angle Deviation Ratio=0.1
• 40 nodes, in 1000mx1000m area• Radio range (R)=50,100,150,200,250m• Simulation time 900sec
UNIVERSITY OFSOUTHERN CALIFORNIA
Link Duration (LD) PDFs
• At low speeds (Vmax < 10m/s) link duration has multi-modal distribution for FW and RPGM4– In FW due to geographic restriction of the map
• Nodes moving in same direction have high link duration
• Nodes moving in opposite directions have low link duration
– In RPGM4 due to correlated node movement• Nodes in same group have high link duration
• Nodes in different groups have low link duration
• At higher speeds (Vmax > 10m/s) link duration does not exhibit multi-modal distribution
UNIVERSITY OFSOUTHERN CALIFORNIA
FW modelVmax=5m/s R=250m
Nodes moving in opposite directions
Nodes moving inthe same direction/lane
Multi-modal Distribution of Link Duration for Freeway model at low speeds
UNIVERSITY OFSOUTHERN CALIFORNIA
RPGM w/ 4 groups Vmax=5m/s
R=250m
Nodes in the same group
Nodes in different groups
Multi-modal Distribution of Link Duration for RPGM4 model at low speeds
UNIVERSITY OFSOUTHERN CALIFORNIA
Vmax=30m/sR=250m
RPGM (4 groups)RW
FW
UNIVERSITY OFSOUTHERN CALIFORNIA
Path Duration (PD) PDFs
• At low speeds (Vmax < 10m/s) and for short paths (h2) path duration has multi-modal for FW and RPGM4
• At higher speeds (Vmax > 10m/s) and longer path length (h2) path duration can be reasonably approximated using exponential distribution for RW, FW, MH, RPGM4.
UNIVERSITY OFSOUTHERN CALIFORNIA
FWVmax=5m/sh=1 hop R=250m
Nodes moving in opposite directions
Nodes moving inthe same direction
Multi-modal Distribution of Path Duration for Freeway model at low speeds, low hops
UNIVERSITY OFSOUTHERN CALIFORNIA
RPGM4Vmax=5m/sh=2 hops R=250m Nodes in the same group
Nodes in different groups
Multi-modal Distribution of Path Duration for RPGM4 model at low speeds, low hops
UNIVERSITY OFSOUTHERN CALIFORNIA
100
Vmax=30m/sR=250m
RPGM4RW
FW
h=2 h=4
h=4
UNIVERSITY OFSOUTHERN CALIFORNIA
Exponential Model for Path Duration (PD)
• Let path be the parameter for the exponential PD distribution: PD PDF f(x)= path e- path x
– As path increases average PD decreases (and vice versa)
• Intuitive qualitative analysis:– PD=f(V,h,R); V is relative velocity, h is path hops & R is radio range
– As V increases, average PD decreases, i.e., path increases
– As h increases, average PD decreases, i.e., path increases
– As R increases, average PD increases, i.e., path decreases
UNIVERSITY OFSOUTHERN CALIFORNIA
UNIVERSITY OFSOUTHERN CALIFORNIA
UNIVERSITY OFSOUTHERN CALIFORNIA
UNIVERSITY OFSOUTHERN CALIFORNIA
But, PD PDF f(x)= path e- path x
Exponential Model for PD
UNIVERSITY OFSOUTHERN CALIFORNIA
FWh=4
0
0.05
0.1
0 10 20 30 40 50
Path Duration (sec)
Prob
abili
ty
Exponential
PDRWh=2
0
0.05
0.1
0.15
0.2
0.25
0 10 20 30 40 50
Path Duration (sec)
Pro
babi
lity
Exponential
PD
RGPM4h=4
- Correlation: 94.1-99.8%
Vmax=30m/s
R=250m
0
0.1
0.2
0.3
0.4
0.5
0 10 20
Path Duration (sec)
Prob
abili
ty
Exponential
PD FWh=4
UNIVERSITY OFSOUTHERN CALIFORNIA
0
0.1
0.20.3
0.4
0.5
0.6
0.70.8
0.9
1
0 2 4 6 8 10 12 14 16 18 20 22 24 26 28 30
Cumulative Distribution Function (CDF)
Pro
ba
bil
ity
Exponential
PDD= 0.048
00.10.20.30.40.50.60.70.80.9
1
0 3 6 9 12 15 18 21 24 27 30 33 36 39 42 45 48 51 54
Cumulative Distrbution Function (CDF)
Pro
ba
bil
ity
Exponential
PDD= 0.0477
0
0.1
0.20.3
0.4
0.5
0.6
0.70.8
0.9
1
0 3 6 9 12 15 18 21 24 27 30 33 36 39 42 45 48
Cumulative Distribution Function (CDF)
Pro
ba
bil
ity
Exponential
PD
D= 0.093
Vmax=30m/s
R=250m - Goodness-of-fit Test
RGPM4h=4
FWh=4
RWh=2
K-S testRW 0.04-0.065FW 0.045-0.085RPGM 0.09-0.12
UNIVERSITY OFSOUTHERN CALIFORNIA
Effect of Path Duration (PD) on Performance (in-progress): Case Study for DSR
• PD observed to have significant effect on performance• (I) Throughput: First order model
– T: simulation time, D: data transferred, Tflow: data transfer time, Trepair: total path repair time, trepair: av. path repair time, f: path break frequency
T
DThroughput
TPD
tTTftTTTT repairflowrepairflowrepairflow .1
... )1(
PD
t
TT
repair
flow
)1
(PD
Throughput
ratePD
t
T
D
PD
tThroughput repair
flow
repair ).1()1(
UNIVERSITY OFSOUTHERN CALIFORNIA
• (II) Overhead: First order model– Number of DSR route requests=– p: non-propagating cache hit ratio, N: number of nodes
• Evaluation through NS-2 simulations for DSR
– RPGM exhibits low , due to relatively low path changes/route requests
Effect of PD on Performance (contd.)
PD
T
PDOverhead
1
Random Waypoint (RW) Freeway (FW) Manhattan (MH)Throughput -0.9165 -0.9597 -0.9132Overhead 0.9753 0.9812 0.9978
Pearson coefficient of correlation () with PD
1
UNIVERSITY OFSOUTHERN CALIFORNIA
Conclusions• Detailed statistical analysis of link and path duration for
multiple mobility models (RW,FW,MH,RPGM4):– Link Duration: multi-modal FW and RPGM4 at low speeds– Path Duration PDF:
• Multi-modal FW and RPGM4 at low speeds and hop count• Exponential-like at high speeds & med/high hop count for all models
• Developed parametrized exponential model for PD PDF, as function of relative velocity V, hop count h and radio range R
• Proposed simple analytical models for throughput & overhead that show strong correlation with reciprocal of average PD
UNIVERSITY OFSOUTHERN CALIFORNIA
Future Work
• Apply path duration analysis to various ad hoc protocols• Attempt to explain: Why does path duration distribution
become exponential?• Analyze convergence time and cache performance to
account for varying performance between different protocols. Use it to extend first order models.
• Analysis of effects of mobility on protocol mechanisms• Extend and release the IMPORTANT mobility tool:
– URL: http://nile.usc.edu/important
UNIVERSITY OFSOUTHERN CALIFORNIA
Backup/Extra Slides
UNIVERSITY OFSOUTHERN CALIFORNIA
Rel vel. (V) prop to max vel (Vmax)
UNIVERSITY OFSOUTHERN CALIFORNIA
D: total data transferred during simulationT: simulation timeTflow: data transfer timeTrepair: total time spent in repairing paths trepair: time to repair path each timePD: average path durationf: frequency of path breakage,r: is constant data rate=D/Tflow
N: number of of nodes
UNIVERSITY OFSOUTHERN CALIFORNIA
• Number of path repairs=T/(PD+trepair)
ratetPD
PD
T
D
tPD
tThroughput
repairflowrepair
repair ).()1(
UNIVERSITY OFSOUTHERN CALIFORNIA
• Expression for cache hit ratio (Hit): [FW model]
Hit=
UNIVERSITY OFSOUTHERN CALIFORNIA
Non-propagating Cache Hit Ratio in DSR (independent of velocity!)