1. A High-Throughput Path Metric for Multi-Hop Wireless Routing Douglas S. J. De Couto, Daniel Aguayo, John Bicket, Robert Morris MIT CSAIL Presented by Valentin Pistol, Duke University April 6 th, 2011 Credits: Includes slides borrowed and edited from authors http://pdos.lcs.mit.edu/grid

2. Test-bed map (indoor wireless network) 2

3. Motivation: Min hop-count not best 3 ‘Best’ for each pair is highest measured throughput of 10 promising static routes.

4. Throughput differs between paths 4

5. Outline 5  Testbed throughput challenges  Wireless routing challenges  ETX - New high-throughput metric  Evaluation

6. Wireless routing challenges 6  Longer routes have less throughput  Many lossy links  Many asymmetric links

7. Challenge: more hops, less throughput 7  Links in a route share radio spectrum  Extra hops reduce throughput Throughput = 1 Throughput = 1/2 Throughput = 1/3

8. Challenge: many links are lossy 8 ‘Good’ ‘Bad’

9. Challenge: many links are asymmetric 9 Broadcast delivery ratios in both link directions. Very asymmetric link.

10. Delivery Ratios 1 mW vs 30 mW 10

11. Metric: Maximize bottleneck throughput 11 Bottleneck throughput: 50%Delivery ratio = 100% 51% D A B C Actual throughput: A-B-C : ABBABBABB = 33% A-D-C : AADDAADD = 25% A-B-C = 50% A-D-C = 51%

12. Metric: Maximize end-to-end delivery ratio 12 A-B-C = 51% A-C = 50% 51%100% 50% A B C Actual throughput: A-B-C : ABBABBABB = 33% A-C : AAAAAAAA = 50% End-to-end delivery ratio:

13. New high-throughput metric (ETX) 13  ETX = expected transmission count  Finding the highest throughput path  Link throughput  1/ Link ETX Delivery Ratio 100% 50% 33% Throughput 100% 50% 33% Link ETX 1 2 3

14. Calculating link ETX 14  Actual TX  P(TX success) = P(Data success) x P(ACK success)  Link ETX = 1/ P(TX success) = 1/ [P(Data success) x P(ACK success)]  Estimating link ETX  Forward delivery ratio d fwd = P(Data success)  Reverse delivery ratio d rev = P(ACK success)  Link ETX  1 / ( d fwd x d rev )

15. Delivery ratios 15  Measured using dedicated link probe packets  All nodes broadcast link probes (134 bytes)  Each node remembers probes for last 10 seconds  Estimates  d fwd  # packets recv / # packets sent  d rev piggybacked on probes from neighbors

16. Route ETX 16  Route ETX = Sum of link ETXs Route ETX 1 2 2 3 5 Throughput 100% 50% 33% 20%

17. ETX characteristics 17  Predicts throughput for short routes  Quantifies loss  Quantifies asymmetry  Quantifies throughput reduction of longer routes  Tends to minimize spectrum use which should maximize system capacity  May reduce energy/packet

18. ETX limitations 18  Does not specifically account for mobility  Link probes susceptible to MAC unfairness and hidden terminals  Route ETX measurements change under load  Estimates based on single small link probe packet (136 bytes)  Under-estimates data loss ratios  Over-estimates ACK loss ratios (38 bytes)  Assumes same bit-rate across all nodes  Assumes loss ratios cannot be controlled by the system

19. Evaluation Setup 19  Indoor network, 802.11b, ‘ ad hoc ’ mode  1 Mbps, 1 mW, small packets (134 bytes), RTS/CTS disabled  DSDV + modifications to metrics  DSR + modifications to metrics

20. ETX improves DSDV throughput 20 DSDV overhead ‘Best’ DSDV+ETX DSDV+hop-count

21. ETX slightly improves on DSR 21 DSR+ETX ‘Best’ DSR+hop-count

22. ETX + DSR (No TX feedback) 22 DSR+ETX DSR+hop-count ‘Best’

23. Summary 23  ETX is a new route metric for multi-hop wireless networks  ETX does find better routes  ETX considers  Throughput reduction of extra hops  Lossy and asymmetric links  Link-layer acknowledgements  DSR + ETX almost identical to DSR with link layer feedback  Feedback already helps avoid links with high loss ratios  No comparison between power levels for same runs

24. Thank you! Questions? 24 image © mrjanitor1.com

25. Backup slides 25

26. Large packets 26

27. Delivery ratio decreases for large packets 27

28. ETX vs. link handshaking 28

29. Hop-count penalty 29

30. ETX vs min hop-count at 30 mW 30