Comparison of Routing Metrics for Static Multihop Wireless Networks

1
Comparison of Routing Metrics for Static
Multi-hop Wireless Networks

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

2
Outline

• Introduction
• Static wireless networks
• Routing background
• Description of four routing metrics
• Evaluation of routing metrics
• Evaluation on a mobile scenario
• Conclusion

3
Introduction

• Primary focus in ad hoc environments is to
  provide scalable routing in the presence of
  mobile nodes
• Increased appearance of applications like
  community wireless networks where most of the
  nodes are stationary
• In static ad hoc networks focus of routing
  changes to improving network capacity or
  performance of individual transfers

4
Characteristics of Static Ad hoc Networks

• Stationary or minimally mobile nodes
• Network topology and links rarely change
• Nodes have unlimited power supply
• Link quality and transmission rates vary

5
Routing basics

• Routing protocols use efficient algorithms for
  minimum cost path selection (Dijkstra,
  Bellman-Ford)
• Costs need to be assigned to links between nodes
• A routing metric is required to calculate said
  costs
• In order for minimum cost path to be computed by
  efficient algorithms the routing metric must be
  isotonic !

6
Isotonicity

• Isotonicity is a sufficient and necessary
  condition for both Bellman-Ford and Dijkstras
  algorithm to find minimum cost paths
• If Dijkstras algorithm is used in hop-by-hop
  routing, strict isotonicity is a necessary
  condition for loop-free forwarding.

7
Ad hoc routing protocols

• DSR, DSDV, AODV use hop count as a routing metric
• Efficient for mobile nodes where topology changes
  fast and quick metric reaction is needed
• Low performance for static mesh networks
• Performance decreases as path length increases
• Does not take link quality into consideration

8
Routing metrics Dos and Donts

• Consider link loss ratio
• Consider link rate
• Capture contention for channel
• Maximize throughput

• Interfere with itself
• Be overly sensitive to load

9
Hop count

• Finds minimum hop paths
• Is simple and fast to compute
• Does not take packet loss or bandwidth of the
  path into account
• Will prefer one-hop lossy path to a two-hop
  reliable path

10
Per-hop Round trip time (RTT)

• Calculates round trip delay seen by probes
  between neighboring nodes
• Send a timestamped probe to all neighbors each
  500 ms
• Receive probe ack from neighbors echoing the
  timestamp and calculate RTT
• Keep a weighted moving average of RTT samples
• Increase average by 20 whenever probe or data
  packet is lost
• Load dependent

11
RTT pros and cons

• Captures several facets of link quality
• If node or neighbor is busy the probe will
  experience high delay
• If other nodes in the vicinity are busy the probe
  will be delayed due to channel contention
• If link is lossy the probe will need to be
  retransmitted several times
• Designed to avoid highly loaded or lossy links
• Queuing delay distorts the RTT values on a hop
• Overhead caused by small probe packets
• Link data rate is not captured by small packets

12
Per-hop Packet Pair Delay (PktPair)

• Calculates the delay between pair of back-to-back
  probes to neighboring node
• Send 2 Packets, 1 small and one large
• Receiving node answers with probe containing
  delay between reception of probes
• Measures several aspects of link quality
• If second probe requires retransmissions by
  802.11 ARQ, delay will increase
• If link has low bandwidth the second probe will
  take more time to traverse the link
• If there is traffic in the vicinity, contention
  will cause probes to delay

13
PktPair pros and cons

• Not affected by queuing delay
• More sensitive to link bandwidth due to large
  probe
• Additional overhead
• Probes are contending with the data packets
  causing self-interference

14
Expected Transmission Count (ETX)

• Presented by De Couto et al.
• Measures the loss rate of broadcast packets
  between neighboring nodes
• Finds high throughput paths minimizing the number
  of necessary transmissions
• where

15
ETX pros and cons

• Broadcasting probe packets reduces overhead
• Suffers little from self-interference because it
  does not measure delays
• Accounts for asymmetric link loss ratios
• Broadcast probes are small and sent at lowest
  possible rate
• Does not directly account for link load or data
  rate
• A heavily loaded link may have low loss rate
• Two links with different data rates may have same
  loss rate

16
Testbed

• 23 nodes in an office environment
• Routing with LQSR (modified version of DSR in
  order to measure link quality)
• Measurements taken during long lived TCP
  connections
• y

17
Median path number and length
18
Throughput in multi-hop pathsHOP vs. ETX
19
Median throughput in multi-hop pathsRTT vs.
PktPair
20
Median Throughput of 300 TCP transfers
21
Comparison of ETX and HOP in a mobile scenario

• Different challenges than in static scenario
• As the sender moves around the network ETX cant
  react fast enough to track the changes in link
  quality
• HOP uses new links as soon as it discovers them

22
Conclusions and further discussion

• In a 23 node static ad hoc network in office
  environment
• RTT and PktPair perform poorly due to load
  sensitivity and self-interference
• ETX outperforms hop count
• In a mobile scenario
• Hop count is more effective because it reacts
  fast to topology changes
• Issues not addressed
• Multi-channel networks
• Capture channel diversity
• Minimize interference
• Load balancing
• Load dependence is not necessarily bad
• Minimize the impact a new flow has on network
  resources