1. Signal-Strength-Aware Routing in Ad hoc Networks Abhinav Gupta Ian Wormsbecker Carey Williamson Dept. of Computer Science University of Calgary

2. October 5, 2004MASCOTS 20042 Ad hoc Networks  An ad hoc network is dynamically formed when two or more mobile hosts with wireless capability come into transmission range of each other  Advantage of ad hoc networks: Can be set up ‘on-the-fly’ Requires no existing infrastructure

3. October 5, 2004MASCOTS 20043 Ad hoc Network

4. October 5, 2004MASCOTS 20044 Objectives  Implementation of Signal-strength- aware AODV  Study the effect of AODV routing, user mobility, and number of hops on TCP throughput.  Study the effectiveness of rate- based-pacing of TCP (TCP-RBP)

5. October 5, 2004MASCOTS 20045 AODV Operation Source Destination RREQ RREP Data RERR Data

6. October 5, 2004MASCOTS 20046 Ad hoc On-Demand Distance Vector Routing Protocol (AODV)  Reactive in operation  Route discovery and maintenance using control packets (RREQ, RREP, RERR and HELLO)  The route freshness determined using sequence numbers associated with the control packets.

7. October 5, 2004MASCOTS 20047 AODV UU  Implementation carried out in user-space.  Consists of 3 kernel modules and some user space modules.  Uses Netfilter hooks for packet mangling from kernel space to user space.  Packets analyzed in user-space to trigger AODV events.  Packets are queued on to user-space using libipq which communicates with ip_queue, standard queue handler for IPv4.

8. October 5, 2004MASCOTS 20048 AODV Modules

9. October 5, 2004MASCOTS 20049 Signal-strength-aware AODV  Rationale: Don’t allow route freshness to be determined solely on the basis of sequence numbers  Checks signal strength of control packets coming from adjacent Mobile Host to before it creates, updates or deletes routes.  Signal strength of control packets determined by link_strength module and used by aodv_socket to choose whether or not to let the packet through.

10. October 5, 2004MASCOTS 200410 Variation of Signal Strength

11. October 5, 2004MASCOTS 200411 Operation  If the packet is a broadcast or is intended for the current host, then it is handled as usual by Linux.  If the packet is not intended for the current host and a route exists, it is forwarded to the next hop.  If no route exists, the packet is dropped.  If the packet is generated by the local host, it is buffered in user-space, and a route discovery initiated and routed to next hop when a route is found.

12. October 5, 2004MASCOTS 200412 Locally Generated Packet NF_IP_LOCAL_OUT Packet

13. October 5, 2004MASCOTS 200413 Experimental Environment  Redhat Linux 8.0 (kernel 2.4.18-14)  5 laptops 3 IBM Thinkpads (Processor P4) 2 Compaq Evos (Processor P3)  Cisco Aironet 350 PCMCIA Wireless Cards  Netperf and Netserver used as source and sink for TCP traffic.  Kernel Probes added to TCP code to monitor the TCP statistics  Airopeek Sniffer used to count the control packets

14. October 5, 2004MASCOTS 200414 Testbed Topology Node1 Running netserver Node2 Node3 Node4 Node5 Running netperf Airopeek Sniffer

15. October 5, 2004MASCOTS 200415 Routing Discovery Time (ms) HopsMinMedianMaxMeanStdDev 23796550.8173.6 36103,212292.8633.9 493315,183613.3999.6

16. October 5, 2004MASCOTS 200416 Round Trip Time

17. October 5, 2004MASCOTS 200417 TCP Throughput (Mbps) HopsStationarySlowMediumFast 22.232.402.371.96 31.452.171.951.66 41.241.591.311.18 Slow Speed – 0.33 m/s Fast Speed – 1.0m/s

18. October 5, 2004MASCOTS 200418 Routing Overhead HopsData PacketsControl PacketsOverhead Ratio 224,3693470.0142394 24,8223490.0140601 24,5163610.0147251 24,5593580.0145771 24,3843490.0143127 317,3694050.0233174 16,9113760.0222340 16,8253760.0223477 17,0723980.0233130 17,1513760.0219229 413,2494990.0376632 12,8574760.0370226 11,9225070.0425264 14,0754520.0321137 13,3244010.0300961

19. October 5, 2004MASCOTS 200419 TCP rate-based-pacing (TCP RBP)  Rationale: “Spread-out” the TCP Packets in time to improve TCP performance  InterPacketDelay=RTT/(CurrentWindow+V)  Performed simulations and experiments with Reno TCP and RBP TCP Related Work: [1] Z.Fu et al, “The Impact of Multi-hop Wireless Channel on TCP Throughput and Loss”, Proceedings of IEEE INFOCOM’03, San Francisco, April 2003 [2] J.Ke and C.Williamson, “Towards a Rate-Based TCP Protocol for the Web”, Proceedings of MASCOTS 2000, San Francisco, pp. 36-45, October 2000

20. October 5, 2004MASCOTS 200420 TCP rate-based-pacing Packet

21. October 5, 2004MASCOTS 200421 Simulations Results (TCP-RBP)

22. October 5, 2004MASCOTS 200422 Experimental Results Num HopsTCP RenoRBP TCP MeanSDevMeanSDev 14.690.024.210.22 22.220.062.130.03 31.440.041.310.11 41.240.011.080.03 TCP Throughput (Mbps)

23. October 5, 2004MASCOTS 200423 Conclusions  Signal-strength-aware AODV is a good choice for ad hoc networks because of low overhead and good performance  Design choices made for signal-strength- aware AODV were effective  Performance of RBP TCP is highly sensitive to channel contention and AODV routing dynamics  Simulation results should be interpreted with caution, unless validated against experimental measurements

24. Thank You!

25. October 5, 2004MASCOTS 200425 Contributions  Provides a working implementation of signal-strength-aware AODV  Demonstrates the functionality and performance of the protocol  Presents empirical measurements of TCP throughput in multi-hop wireless ad hoc network  Evaluates the effectiveness of TCP rate- based-pacing in an experimental setting