Comparison of Routing Metrics for a Static Multi-Hop Wireless Network

1
Comparison of Routing Metrics for a Static
Multi-Hop Wireless Network

• Richard Draves, Jitendra Padhye, Brian Zill
• Microsoft Research

Presented by Jón T. Grétarsson
CS577 Advanced Computer Networks
2
Outline

• Introduction
• Setup
• Results
• Conclusions
• Discussion

CS577 Advanced Computer Networks
3
Introduction
CS577 Advanced Computer Networks
4
The Problem

• In recent years, ad hoc wireless networks have
  emerged as a hot topic
• Started with Military Applications
• Commercial Applications of multi-hop wireless
  networks becoming popular (Roofnet, BAWUG,
  Seattle Wireless)
• Quality of links arent taken into account in
  current routing algorithms

CS577 Advanced Computer Networks
5
The Authors

• Richard Draves
• Jitendra Padhye
• Brian Zill

CS577 Advanced Computer Networks
6
The Paper

• About Routing Metrics in Mesh Networks
• Presented in ACM SIGCOMM, 2004
• A summary for the impatient

CS577 Advanced Computer Networks
7
Setup
CS577 Advanced Computer Networks
8
The Metrics

• Hop Count (HOP)
• Per-hop Round Trip Time (RTT)
• Per-hop Packet Pair Delay (PktPair)
• Expected Transmission Count (ETX)

CS577 Advanced Computer Networks
9
Ad Hoc Routing Architecture

• Mesh Connectivity Layer
• Layer 2.5 Architecture
• Link Quality Source Routing

CS577 Advanced Computer Networks
10
LQSR

• Modified DSR to include Link Quality Metrics
• Link-State routing

CS577 Advanced Computer Networks
11
Testbed
CS577 Advanced Computer Networks
12
Testbed

• 23 Nodes
• Not Wireless-Friendly
• High Node Density
• Wide Variety of Multi-Hop Paths
• 801.11a Wireless Network
• Static Positions

CS577 Advanced Computer Networks
13
Results
CS577 Advanced Computer Networks
14
LQSR Overhead

• CPU Bottleneck for shorter paths
• Channel Contention for longer paths

CS577 Advanced Computer Networks
15
Link Variability

• 183 of 506 Links displayed activity

CS577 Advanced Computer Networks
16
Link Variability

• 90 Links with non-zero bandwidth in both
  directions

CS577 Advanced Computer Networks
17
Long Lived TCP Flows

• Transfer duration fixed
• One active transfer at a time
• Semi-Inter Quartile Range bars
• Large variations in throughput
• UDP vs TCP
• Self-Interference

CS577 Advanced Computer Networks
18
Median Throughput
CS577 Advanced Computer Networks
19
Median Number of Paths
CS577 Advanced Computer Networks
20
Path Length

• As path length increases, throughput decays
• Testbed diameter is 6 7 hops
• Self-Interference is still a big problem for RTT
  and PktPair
• ETX appears to approach a non-zero asymptote

CS577 Advanced Computer Networks
21
Median Path Length
CS577 Advanced Computer Networks
22
Average Path of ETX vs HOP
CS577 Advanced Computer Networks
23
RTT Throughput vs Path Length
CS577 Advanced Computer Networks
24
PktPair Throughput vs Path Length
CS577 Advanced Computer Networks
25
HOP Throughput vs Path Length
CS577 Advanced Computer Networks
26
EXT Throughput vs Path Length
CS577 Advanced Computer Networks
27
Variability of Throughput

• Coefficient of Variation
• 6 periphery nodes to 5 receivers
• 1 active transfer at any time

CS577 Advanced Computer Networks
28
Median Throughput
CS577 Advanced Computer Networks
29
CoV of ETX vs HOP
CS577 Advanced Computer Networks
30
Competing TCP Transfers

• RTT not worth demonstrating
• Multiple Median Throughput (MMT)

CS577 Advanced Computer Networks
31
Competing TCP Transfers
CS577 Advanced Computer Networks
32
Web Traffic

• Only one client active at any time
• 1300 files fetched
• Transfer using Surge
• File size within the range 77B, 700KB
• Measured latency

CS577 Advanced Computer Networks
33
Median Overall Latency
CS577 Advanced Computer Networks
34
Median Latency lt1KB
CS577 Advanced Computer Networks
35
Median Latency gt8KB
CS577 Advanced Computer Networks
36
Web Traffic Conclusions

• In longer paths, ETX dominates
• In shorter paths, HOP sometimes wins

CS577 Advanced Computer Networks
37
Mobile Scenario
CS577 Advanced Computer Networks
38
Mobile Results
CS577 Advanced Computer Networks
39
Mobile Results

• ETX has problems adjusting quickly enough
• HOP has no such problems

CS577 Advanced Computer Networks
40
Conclusions
41
Paper Conclusions

• RTT and PktPair are load-sensitive and suffer
  from Self-Interference
• ETX significantly outperforms HOP in the
  stationary ad hoc network
• ETX relative performance gain increases as path
  length increases
• HOP responds faster to the changes of a mobile ad
  hoc network

CS577 Advanced Computer Networks
42
Discussion
43
Discussion

• Experimental Flaws
• Logical Fallacies
• Beating Up competition
• What didnt the authors do?

CS577 Advanced Computer Networks