Multihop Wireless Networks: What’s Wrong With Min Hopcount? Douglas S. J. De Couto Daniel Aguayo, Benjamin A. Chambers, and Robert Morris http://pdos.lcs.mit.edu/grid 1. NOT shortest path is bad, but we need to be more discerning

Wired Nets: Why Min Hopcount? Prefer route with fewest links (“hops”) Many more sophisticated alternatives End-to-end latency, lowest congestion, etc. Min hopcount works in practice Can engineer a decent network Alternatives complex, less tested

Min Hopcount Assumption All important route state in hopcount Link quality is bimodal Links are either “good”, or “bad” All “good” links are equivalent Sufficient condition for success What about wireless?

Indoor Wireless Network 18 nodes 6 on 6th, 12 on 5th 802.11b radios (fixed tx power), PCs, DSDV 5th floor 6th floor

Min Hopcount Underperforms Packets/sec Worse Better Why DSDV, doesn’t it suck? No, not on static nets. Max 802.11 small packet throughput is ~400—750 pps, depending on bit rate DSDV worse on 152/185 pairs 18 * 17 = 296 total possible pairs; no data for some of the pairs (either missing DSDV or SR) Best half of SR routes are mostly 1-hop with a few scattered 2-hop. The middle third and bottom are mostly 2-hop, with a few 3-hop among the bottom half. 124 bytes * 500 pkts/sec ~= kilobits/sec (kbps) Why best sometimes not better: experiments at different times, didn’t try all possible routes Node Pair DSDV 124-byte packets for 30 seconds “Best” static route “Best” is maximum pkts/sec over multiple static source routes

Not All Min Hopcount Paths Are Equal 3-hop paths between a single node pair 14 paths, all shortest paths between same 2 nodes: 3 hops Different experiment from other slides, same packet size: 124 bytes Different throughput for different paths! Pair: 10 to 18

Not All Links Are Equal 171 non-zero links Links are not bimodal Broadcast delivery rate Link 171 non-zero out of 342 links measured. But why not 18*17=306 links? Oops! In data collection… 171 non-zero links Links are not bimodal Min hopcount assumptions are false

Intermediate Quality Wireless Links Always Exist In a random ad hoc network, there will be nodes at all distances from each other Since link quality depends on signal strength, and signal strength is a function of distance, there will nodes with all different link qualities from any given node. Min hopcount -> Max distance per hop -> min hopcount more likely to pick marginal links! There are nodes at every distance Link quality depends on signal strength (distance) Min hopcount maximizes link distance Marginal links are more likely!

Should We Give Up Hopcount? Intuition from wired networks is wrong Links share spectrum Capacity penalty for more hops What should we optimize? Per-route throughput, network capacity, power? Capacity = 1/2 Capacity = 1

Low Quality 802.11 Links 802.11 has link-level retransmissions Lossy links become narrow links 0% effective loss rate Good pings But, low packet-per-second throughput Retransmissions waste capacity Other nodes could have transmitted Will retransmit until ACK, up until a fixed number of maximum retransmits, e.g. 16

A Reasonable (Wrong) Idea Maximize bottleneck throughput: A-D-C 50% 100% 51% D A B C A-B-C: ABBABBABB = 1/3 A-D-C: AADDAADD = 1/4 Actual throughput:

A Better Idea Insight: spectrum use is important! Idea: minimize transmissions per packet Num tx 1 2 2

Transmissions Per Packet B 100%1 tx 50% 2 tx A C 51% 2 tx 51% 2 tx D Transmissions per packet (A-B-C) = 3 Transmissions per packet (A-D-C) = 4

Research Agenda Explore performance of other protocols, e.g. DSR, AODV Explain route performance by underlying link performance Confirm usefulness of transmissions per packet as metric Handle link variation over time

Summary Minimum hopcount protocols are unlikely to achieve best performance Quality varies between links Transmissions per packet nicely quantifies link and path quality for routing