Routing in Multi-Radio, Multi-Hop Wireless Mesh Networks CS598JH Routing in Multi-Radio, Multi-Hop Wireless Mesh Networks Richard Draves, Jitendra Padhye, and Brian Zill Microsoft Research Presented by Zahid Anwar

Overview Core Idea and Contributions of Paper Design of the framework and WCETT metric Experimental Results Cons

Background Shortest-path is not always the best path Link characteristic is not ON-OFF ETX metric for a link = expected number of retransmissions on that link Successful packet transmission = DATA + ACK Measure packet loss rate pf, pr using broadcast packets Then, ETX = 1/[1 – (1- pf).(1-pr)] ETX metric for a path = sum of ETX metrics for links along the path lossy link good link

Core Idea ETX and multiple radios ETX does not consider bandwidth while selecting paths, so it will choose 802.11b over 802.11a if the loss rates are the same (longer range 802.11b links). ETX does not give any preference to channel-diverse paths (more on this later) Idea: if we use multiple radios at every node, node can simultaneously transmit and receive node can simultaneously transmit on multiple channels “self-interference” in routes can be reduced can improve robustness to channel variation, noise by using different parts of the spectrum (802.11a/b)

Contributions New routing metric for multi-radio mesh networks Weighted Cumulative Expected Transmission Time (WCETT) Implementation of the metric in a link-state routing protocol called Multi-Radio Link-Quality source routing (MR-LQSR) Evaluation on a 24-node, multi-radio mesh testbed WCETT makes judicious use of two radios Over 250% better than HOP Over 80% better than ETX Gains more prominent over shorter paths and in lightly-loaded scenarios

Multi-Hop Networks with Single Radio With a single radio, a node can not transmit and receive simultaneously.

Multi-Hop Networks with Multiple Radios With two radios tuned to non-interfering channels, a node can transmit and receive simultaneously. Increased robustness due to frequency diversity e.g. 2.4GHz (802.11b) and 5GHz (802.11a) have different fading characteristics Possible tradeoff between range and data rate Advantages of multiple radios for improving capacity are + allows simultaneous transmit and receive (otherwise with only one radio the capacity of relay nodes is halved) + the network can utilize more of the radio spectrum + radios operate on different freq bands have different bandwidth, range and fading characteristics + radios are off the shelf commodity parts with diminishing prices

Design of Routing Metric Model Nodes have one or more radios tuned to non-interfering channels No power constraints Little or no node mobility (Relatively stable links) Design goals for a new path metric Takes bandwidth and loss rates on each link into account Adding a link to the path does not decrease the path metric Explicitly accounts for reduction in throughput due to interference among links that operate on the same channel

Implementation Framework Implemented in a source-routed, link-state protocol Multi-Radio Link Quality Source Routing (MR-LQSR) Node discovers links to its neighbors; Measure quality of those links Link information floods through the network Each node has “full knowledge” of the topology Sender selects “best path” Packets are source routed using this path

Components of a Routing Metric Link Metric: Assign a weight to each link Path Metric: Combine metrics of links on path ETX: Prefer low-loss links WCETT: Prefer high-bandwidth, low-loss links HOP: Each link has weight 1 WCETT: Prefer short, channel-diverse paths HOP: Path Metric = Sum of Link Metrics ETX: Prefer short, low-loss paths

Link Metric: Expected Transmission Time (ETT) Link loss rate = p Expected number of transmissions Packet size = S, Link bandwidth = B Each transmission lasts for S/B Lower ETT implies better link ETX formula bigger, no highlight

ETT: Illustration ETT : 0.77 ms ETT : 0.89 ms ETT : 0.77 ms 11 Mbps 5% loss 18 Mbps 50% loss 18 Mbps 10% loss 1000 Byte Packet ETT : 0.77 ms ETT : 0.89 ms 1000 Byte Packet ETT : 0.77 ms ETT : 0.40ms

Combining Link Metric into Path Metric Proposal 1 Add ETTs of all links on the path Use the sum as path metric SETT = Sum of ETTs of links on path (Lower SETT implies better path) Pro: Favors short paths Con: Does not favor channel diversity

SETT does not favor channel diversity 6 Mbps No Loss 6 Mbps No Loss 1.33ms 1.33ms 1.33ms 1.33ms 6 Mbps No Loss 6 Mbps No Loss Path Throughput SETT Red-Blue 6 Mbps 2.66 ms Red-Red 3 Mbps 2.66 ms

Impact of Interference Interference reduces throughput Throughput of a path is lower if many links are on the same channel Path metric should be worse for non-diverse paths Assumption: All links that are on the same channel interfere with one another Pessimistic for long paths

Combining Link Metric into Path Metric Proposal 2 Group links on a path according to channel Links on same channel interfere Add ETTs of links in each group Find the group with largest sum. This is the “bottleneck” group Too many links, or links with high ETT (“poor quality” links) Use this largest sum as the path metric Lower value implies better path “Bottleneck Group ETT” (BG-ETT)

BG-ETT Example BG-ETT favors high-throughput, channel-diverse paths. 6 Mbps 6 Mbps 6 Mbps 6 Mbps 6 Mbps 6 Mbps 1.33 ms 1.33 ms 1.33 ms 1.33 ms 1.33 ms 1.33 ms Path Throughput Blue Sum Red Sum BG-ETT All Red 1.5 Mbps 5.33 ms 1 Blue 2 Mbps 1.33 ms 4 ms Red-Blue 3 Mbps 2.66 ms BG-ETT favors high-throughput, channel-diverse paths.

BG-ETT does not favor short paths 6 Mbps 6 Mbps D 6 Mbps 1.33 ms 1.33 ms 1.33 ms 6 Mbps 1.33 ms 4 ms 2 Mbps S D Path Throughput Blue Sum Red Sum BG-ETT 3-Hop 2 Mbps 4 ms 4-Hop 2 Mbps 4 ms

Path Metric: Putting it all together SETT favors short paths BG-ETT favors channel diverse paths Weighted Cumulative ETT (WCETT) WCETT = (1-β) * SETT + β * BG-ETT β is a tunable parameter Higher value: More preference to channel diversity Lower value: More preference to shorter paths

How to measure loss rate and bandwidth? Loss rate measured using broadcast probes Similar to ETX Updated every second Bandwidth estimated using periodic packet-pairs Updated every 5 minutes

Mesh Testbed 23 nodes running Windows XP. Two 802.11a/b/g cards per node: Proxim and NetGear (Autorate) Diameter: 6-7 hops.

Experiment Setting 2-Minute TCP transfer between 100 randomly selected node pairs (Out of 23x22 = 506) Only one transfer active at a time Performance metric: Median throughput of 100 transfers Baseline (Single Radio) NetGear on 802.11a (Channel 36), Proxim OFF Two Radio NetGear on 802.11a (Chan 36), Proxim on 802.11g (Chan 10) (802.11g radios have longer range, lower bandwidth)

Median Throughput (Baseline, single radio) Increase in performance is a result of the fact that wcett takes link bandwidth into account This sometimes leads it to select longer paths than ETX, however these longer path result in better throughput WCETT provides performance gain even with one radio.

Median Throughput (Two radios) Outperformed ETX because it doesn’t account for bandwidth or channel diversity WCETT takes better advantage of the additional capacity provided by the 2nd radio Shortest –path simply selects 802.11g which have lower available bandwidth than 802.11a WCETT makes judicious use of 2nd radio: 86% gain over baseline Performance of HOP worsens with 2nd radio! ETX can not take full advantage of 2nd radio

Do all paths benefit equally with WCETT? Group connections by their path length, and plot the improvement in their media throughput, when compared to their throughput in the single radio baseline case Does not provide any improvement for single-hop connections because it does not strip packets over multiple links between neighbouring nodes. For multi-hop connections, the performance improvement drops with increase in path length Surprising because benefits of channel diversity should be more evident on longer paths Authors attribute this to poor performance of TCP WCETT gains are more prominent for shorter paths

Channel diversity is important; especially for shorter paths Impact of β value WCETT metric is a weighted average of 2 quantities: 1st the sum of ETTs of all hops along a path and the 2nd the sum of ETTs on the bottleneck channel. Metric selects paths with less channel-diversity when B is low On paths of length 4 or more, channel diversity does not provide significant benefit Channel diversity is important; especially for shorter paths

Performance of Two Simultaneous Flows 2-Minute TCP transfer between 100 randomly selected node pairs Two transfers active at a time Two radios: Netgear: 36-a, Proxim: 10-g Performance metric: 2 x Median throughput Repeat for ETX and WCETT (β = 0, 0.1, 0.5, 0.9)

Two simultaneous flows β = 0: No weight to diversity β = 0.9: High weight to diversity WCETT performs better than ETX for all values of B in this scenerio At high load levels, the total network throughput is maximized by using lower values of B Throughput better for lower values of β WCETT Performs better than ETX Channel Diversity is important

Cons Captures intra-flow interference but what about inter-flow? Unrealistic Test bed: didn’t consider virtual carrier sensing, > 2 radios scenarios and > 2 flows Authors suggest WCETT might not work with other other routing protocols like AODV How to select β ? Maybe automatic selection of based on load levels

Cons continued WCETT is not isotonic like shortest path and ETX Doesn’t work well with algorithms like Bellman-Ford and Dijkstra's Requires algorithms of exponential complexity Incorrectly chooses S1  S2  C  D  T over S1 B  T Correctly chooses S2  S1  B  T Causes forwarding holes in link state routing protocols

For more information http://research.microsoft.com/mesh/ Source code, binaries, tech reports, …

Backup Slides

Impact of Interference Intuition 1 Impact of interference can be captured by adding ETTs of interfering links 6 Mbps, No Loss 6 Mbps, No Loss 1.33 ms 1.33 ms Path throughput: 3 Mbps Sum of ETTs: 2.66 ms

Why does ETX not do well? ETX does not take bandwidth and channel diversity into account