1 Expected Data Rate (EDR): An Accurate High-Throughput Path Metric For Multi- Hop Wireless Routing Jun Cheol Park Sneha Kumar Kasera School of Computing University of Utah
2 Multi-hop wireless networks Flexible solution regardless of existence of fixed wired infrastructure Efficient ad hoc routing necessary to achieve high throughput Path metric crucial in selecting ad hoc paths
3 Related Work ETX (Expected Transmission Count) [MobiCom’03] considers packet loss, but does not accurately model transmission interference Existing transmission interference models do not consider packet loss None of existing work has comprehensively addressed packet loss, transmission interference together
4 ETX Average # transmissions (including retransmissions) needed for successful packet delivery on wireless link with loss rate p ETX sum of ad hoc path sum of ETX of individual links used as path metric for selecting best ad hoc path Achievable Data Rate of a link: Maximal data rate / ETX Maximal data rate delivery ratio ETX = 1 - p 1
5 Limitations of ETX sum UDP packet size: 1500 bytes Source node always backlogged (11 Mbps) ETX sum cannot accurately differentiate ad hoc paths
6 Goal Develop an accurate high-throughput path metric for multi-hop wireless networks
7 Outline Problem Setting EDR (Expected Data Rate) Transmission Contention Degree Back-off procedure Performance Evaluation Summary
8 Problem Setting IEEE networks Distributed Coordination Function (DCF) all links use single data rate Load-insensitive path metric, routing does not consider “dynamic interference” due to other flows considers “unavoidable” transmission interference within single flow 1234
9 Basic Ideas of EDR Every link relies on supplying rate from previous link EDR : achievable data rate of whole ad hoc path = achievable data rate of bottleneck link B: Bottleneck link D: Maximal Data rate on link B ETX(B) D EDR = ETX(B) for wired links
10 Basic Ideas of EDR Every link relies on supplying rate from previous link EDR : achievable data rate of whole ad hoc path = achievable data rate of bottleneck link B: Bottleneck link D: Maximal Data rate on link B I: Total transmission interference factor ETX(B) D ETX(B) I EDR = for wireless links
11 Total Transmission Interference Factor Depends upon TCD: Transmission Contention Degree RTCD: Relatively Increased TCD I = Sum of all TCD and RTCD on links that interfere with bottleneck link B
12 Transmission Contention Degree for Link k Represents how busy link k transmitting, retransmitting packets range [0.0, 1.0], normalized value compared maximal data rate of link k when node always backlogged, TCD = 1.0 Considers load due to original transmission, retransmissions
13 How to calculate TCD? TCD(k) ETX(k) Supplying rate at link k+1 = TCD(k+1) ? TCD(k) ETX(k) ETX(k+1) ETX(k) TCD(k) TCD(k+1) = Min { 1, ETX(k+1) } Assume ETX values of links are given TCD(k+1) in terms of TCD(k)? TCD(1) = 1.0 Original load Increased load due to lost packets
14 Effect of Back-off No mechanism to differentiate packet loss due to collisions, channel noise Upon packet loss exponential back-off used for occupying shared medium Different loss rates between adjacent links different average contention window sizes different medium occupancy probabilities relatively increased TCD (RTCD) on higher loss rate link
15 How to calculate RTCD? Assume W(1) = 5, W(2) = 10 Node 1 twice more likely to occupy shared medium than Node 2 Thus, higher loss rate node (Node 2) experiences relative increase in TCD due to different window sizes RTCD(k+1) = W(k+1)/W(k) Window size W(k)
16 EDR ETX(B) I EDR = D D: Maximum data rate on bottleneck link B ETX(B): ETX of link B I: Sum of (TCD+ RTCD) over all links that interfere with link B
17 Performance Evaluation NS-2 simulations Independent, temporally correlated loss models Randomly generate 270 ad hoc paths hop lengths: link loss rates: (ETX: ) Construct groups of 4 ad hoc paths between source, destination for given group as input set, find how well each metric selects best ad hoc path Use 1500-byte UDP packets, send rate at source node = 11 Mbps
18 Independent loss EDR performs much better than ETX sum EDR: for 90% of input cases, throughput more than 90 % of best
19 Temporally correlated loss Packet burst loss modeled using two-state continuous time Markov chain Burst length borrowed from experimental results [Divert, MobiSys ’04]
20 Summary Proposed a new metric, EDR Showed that EDR can accurately determine achievable data rates of ad hoc paths Future work investigate TCP over EDR routing apply EDR in multi-radio wireless networks
21 Backup
22 EDR for TCP on multi-rate paths I EDR = R Bottleneck link B such that R = Min { D(k) / ETX(k) } I = TCD(k)/TCD max, k over interference range of link B, Normalized total transmission contention degree in terms of B For TCP flows, EDR does not include RTCD in I because TCP window mechanism is able to avoid unnecessary overhead of RTCD by adjusting send rate at source node ETX(k+1) ETX(k) TCD(k+1) = TCD(k) D(k) D(k+1) , TCD(1) = 1.0