1 Paper: IMPORTANT by Bai Sadagopan et al. Michalis Faloutsos
Feb 05, 2016
1
Paper: IMPORTANTby Bai Sadagopan et al.
Michalis Faloutsos
2
The Problem: How important is mobility in ad hoc performance?
Mobility has been ignored or downplayed so farQuestions:• What is the effect of mobility on performance?• How does mobility affect performance?
3
Contribution: Mobility is critical
Mobility affects the performanceDefine a set of parameters (framework):• Mobility• Network connectivity• Protocol structure (building blocks)
Relative performance of protocols varies with mobilityMobility affects: connectivity (link/path stability)
4
Motivation and Significance
W need a systematic approach to simulationsWarn us: varying scenarios gives diff. ResultsProvide a framework to characterize mobility
5
Roadmap
Previous workDefinition of metrics for mobility and connectivityExperimental resultsConclusionsPaper criticism
6
Mobility models and studies
Way- point modelGroup Mobility (RPGM)• Groups of nodes that follow a group leader• Each member has speed similar to leader:
Magnitude, angle (from x-axis)Freeway: predetermined lanes two wayManhattan: square lattice two-way traffic• Prob. 0.5 go straght, 0.25 turn left, 0.25 turn
right
7
Previous studies
Mobility does matter Pei et al. [11]:Proactive and reactive protocols are similar performance johansson et al. [10]Proactive protocols better throughput, and e2e delay, but reactive protocols less overhead Das et al. [8] (spot any contradiction?)• [8] introduced a mobility model limiting
“change”
8
Metrics: Capturing mobility
9
Metrics: Spatial and Temporal Dependence
Spatial Dependence: movement similarity of nodes close together
Temporal Dependence: similarity of node movement with itself in the near past
Relative Speed: as in physics: difference of vectors
10
Metrics for graph connectivity
Number of link changes: how many times there was a direct link between two nodesLink duration: how long a link between nodes lastsPath availability: fraction of time that two nodes are connected
11
Simulation set up
Area: 1000m x 1000m40 nodes900 secRPGM groups of 1x40, and 4x10Speed: Vmax = 1,5,10,20,30,40,50,60 m/secRange: 250m
12
Which metrics can distinguish mobility?
Average relative speed and spatial dependence are good metrics
13
Metrics for connectivity
Average link duration distinguished mobility models
Why this dip?
14
Mobility affects protocol performance
AODV and DSDV thruput comparison variesProactive (DSDV) protocols sometimes better than reactive
Transmission range 250m
15
Mobility models
What do nodes do when they reach the end of the road?
16
Conclusions
Mobility affects performance of protocolsDSR is a well “designed” protocol with optimized parameters• Other protocols are non-optimized then?
DSR: aggressive caching of routes• Good for low mobility
17
Criticism
Mobility was examined, but what about other parameters:• Average path length of connection• Average duration of connection• Speed with respect to connection and
simulation durationParameters of different protocols, were they optimized?
18
Quick calculations
Relative distance in hops? 1000^2/40 = 25000 m^2Area of transmission: pi r^2 = 3.14 250^2 = 196,250Average neighborhood= 7.85 nodesNode density!Max straight path length: 1000 sqrt(2) / 250 = 5.65 hopsAverage path length: approx 3 hops • Is that “enough”?
19
Mean criticism
We knew that mobility is importantWe knew that DSR seems to perform best• In one case AODV does better slightly
Do their results suggest what is the “right” simulation? No.They provide some intuitive obseervationsBuilding blocks: do they actually explain something?
20
Fair criticism
Provide a framework, first organized attemptProvide metricsDo thorough experimentsExplain they results they see (most of them)Intersting work that makes you think
21
How could I do things better
Model topology from a graph theoretic poin of viewMore metrics: • Average path length• Average neighborhood• Node density• Path length inflation• Choice of source-destination pairs (skewed
distr.)Measure disconnectivity: scope, duration
22
More how to do things better
Examine the effect of other parameters than just mobilityShow that subtle parameters in protocols can make significant changesArgue that protocol parameters must be dynamic!