1. Robust MANET Design John P. Mullen, Ph.D. Timothy I. Matis, Ph.D. Smriti Rangan Karl Adams Center for Stochastic Modeling New Mexico State University May 16, 2004

2. 11/20/20032 What Are MANETS ? A MANET is a mobile ad-hoc wireless communication network that is capable of autonomous operation Each node is capable of transmitting, receiving, and routing packets of information. The network has no fixed backbone (such as with the Internet and cellular phones) The nodes are able to enter, leave, and move around the network independently and randomly

3. 11/20/20033 Mobile Ad Hoc Path Search Y X A B I G E F C D H

4. 11/20/20034 Same MANET After a While Y X A B I G E F C D H H X I G F E D B A C Y

5. 11/20/20035 Types of Packets Control Packets – –RREQ’ s and RREP’s – Used to establish communication links between the source and destination nodes. There are numerous protocols that have been proposed for their “optimal” use in finding reliable links at minimal bandwidth –ACK’s – Used to ascertain the quality of a link and ensure successful communication. The destination node sends an acknowledgement (ack) packet back to the source after each successful data packet transmission. Data Packets –Contain the actual information that is to be communicated broken up into “packets” of uniform size –Data packets are much larger than control packets

6. 11/20/20036 Protocol Taxonomy Single channel protocols uniform Destination based reactive proactive topology-based reactive proactive Non-uniform partitioning Neighbor selection AODV TORA ABR DSDV WRP DSR GSR CEDAR CBRP ZRP OLSR

7. 11/20/20037 MANET Models Current MANET Models –Received power is a deterministic function of distance –Node communication (p received  p min ) is flawless within a nominal range, r 0, and is not possible (p received  p min ) beyond this range In actuality, the received power process is highly stochastic due primarily to shadowing and fading

8. 11/20/20038 Current Assumption: Rec. Power is a deterministic function of distance p(r) Current vs. Observed Field Measurements: From Neskovic 2002 – Fig. 2

9. 11/20/20039 Evaluating Protocols The deterministic power assumption is the default of most simulation software (OpNet, NS2, NAB) used to evaluate protocol performance The stochastic problem is typically viewed as a minor (and unimportant) nuisance by the CS and EE communities that design these protocols

10. 11/20/200310 Rayleigh Fading The instantaneous received voltage in an inefficient, low power, and complex RF environment often follows a Rayleigh distribution As a result, it follows that received power is exponentially distributed Further, we assume power exponentially decays with distance

11. 11/20/200311 PDF of Received Power

12. 11/20/200312 Initial Test Scenario

13. 11/20/200313 Rec Power –Current Model

14. 11/20/200314 Current vs Proposed Model

15. 11/20/200315 Real Vs. Memorex

16. 11/20/200316 Impact on Link Throughput

17. 11/20/200317 Findings Not all packets within nominal range are transmitted successfully Not all packets outside the range are unsuccessful

18. 11/20/200318 Scenario Two – DSR Protocol Source Relay Dest.

19. 11/20/200319 RF Propagation Distances Source Relay Dest.

20. 11/20/200320 Throughput

21. 11/20/200321 End-to-End Delay Delay = 0.004 sec In no-fading model

22. 11/20/200322 Route Discovery Time One Route discovery In no-fading model

23. 11/20/200323 Transmit Buffer Size Buffer size is 2.0 In no-fading model

24. 11/20/200324 Hops per Route 1.5 hops average A-B: 1 hop A-C: 2 hops In no-fading model

25. 11/20/200325 The Basic Problem Source Relay Dest.

26. 11/20/200326 Ping - Pong 2-hop 1-hop 0.6 0.4 ABC ABC 0.8 0.2 0.995 1 - 0.4 6 0.005 0.75 0.25 p 2 = 2p 1 p 2 = 50p 1

27. 11/20/200327 Throughput vs. Tries

28. 11/20/200328 Delay vs. Tries

29. 11/20/200329 Buffer Size vs. Tries

30. 11/20/200330 Findings Only through accurate stochastic simulations can 1.The true performance of existing protocols be evaluated 2.The parameters of these protocols be optimized for robust performance 3.New robust protocols be developed Parameters not significant in deterministic models (such as packet retry) are important in stochastic models

31. 11/20/200331 Robust MANET Design RSM may be used to optimize the performance of established protocols for the controllable parameters (F, number of TX tries, etc.) over the uncontrollable parameters (c, TX rate, etc.) As an example, consider optimizing the number of TX tries (1,2,3,4) over 2 levels of TX rate (71.5,143 in packets/sec) using throughput as a measure of performance

32. 11/20/200332 Throughput (packets/sec)

33. 11/20/200333 Throughput (High/Low Data Rates)

34. 11/20/200334 Relative Throughput

35. 11/20/200335 Relative Throughput (High/Low)

36. 11/20/200336 Mean Delay

37. 11/20/200337 Mean Delay (High/Low)

38. 11/20/200338 Mean Transmit Buffer Size

39. 11/20/200339 Mean Total Bits Per Second

40. 11/20/200340 Mean Routing Bits per Second

41. 11/20/200341 Mean Non-Routing Bits

42. 11/20/200342 Questions ? John Mullen jomullen@nmsu.edu Tim Matis tmatis@nmsu.edu Center for Stochastic Modelling http://engr.nmsu.edu/~csm