1. CSE 802.11 University of Washington Multipath Routing Protocols in AdHoc Networks

2. First Question  Why another lecture on ad-hoc routing protocols?  Keyword: multipath  Ability to provide multiple routing options from a source to a destination  Have “enough redundancy” to handle the ad-hoc nature of these networks  Today, the multipath property of routing protocols will take center stage

3. But … before we begin….  Why do we need different protocols than in the Internet?  The attributes of the problem have changed:  Infrastructureless  Nodes’ availabilities are very low…  Must provide incentives to cooperate  Adequate security story  Cooperation-induced trust  Mobility: topology changes frequently  Power consumption: communication and computation consume energy  Network environment:  More complex (subject to more factors)  Factors change frequently and in unpredictable ways

4. But … before we begin….  Why do we need different protocols than in the Internet?  The attributes of the problem have changed:  Infrastructureless  Nodes’ availabilities are very low…  Must provide incentives to cooperate  Adequate security story  Cooperation-induced trust  Mobility: topology changes frequently  Power consumption: communication and computation consume energy  Network environment:  More complex (subject to more factors)  Factors change frequently and in unpredictable ways P2P!!!

5. 90% of sessions last less than 1 hour

6. 90% of sessions last less than 10 mins

7. 50% of sessions last less than 1 hour

8. The Availability Story Revealed  Consider a path P of 4 hops and 5 nodes  Assuming independent behavior  What is the probability that P will change within the next 10 minutes?  Ad-Hoc: 89%  P2P: 77%  Can’t afford to propagate every change throughout the network (like Internet protocols)  Although you might need node discovery  Need on-demand routing protocols

9. Desirable Characteristics of Ad- Hoc Routing Protocols  Adapt to rapid changes  Multipath  On-demand  Low communication overhead  Low computation overhead  Provides incentive to cooperate  Secure  Loop-free  Maintains node connectivity

10. Desirable Characteristics of Ad- Hoc Routing Protocols  Adapt to rapid changes  Multipath  On-demand  Low communication overhead  Low computation overhead  Provides incentive to cooperate  Secure  Loop-free  Maintains node connectivity

11. Multi-path Ad-Hoc Routing Protocols  Multipath DSR  2 versions in today’s paper  AOMDV  AODV-BR  ROAM  MDVA

12. Multi-path Ad-Hoc Routing Protocols  Multipath DSR  2 versions in today’s paper  AOMDV  AODV-BR  ROAM  MDVA

13. DSR - Route Discovery Example Route Request Message 1 2 3 5 4 7 6 8 9 10 1 1 1, 2 1, 3 1, 2, 5 1, 3, 4 1, 2, 5, 6 1, 3, 4, 7 1, 3, 4, 7, 9 1, 2, 5, 6, 8 Route Reply 1, 3, 4, 7, 9, 10 From: Stefan Dulman (dulman@cs.utwente.nl)

14.  How it works:  S broadcasts a Route Request message to D  Each node forwards request by adding its own address and re-broadcasting  Requests propagate until:  The target is found  A node that knows a route to D is found DSR Route Discovery

15. Route Maintenance  If a node does not receive a confirmation from the next node that the packet was successfully forwarded, initiates a Route Error message back to the source  The data packet will be transmitted over another existing path (if multipath) or a new route discovery could be initiated

16. DSR – With Multipath (1st Version) Route Request Message 1 2 3 5 4 7 6 8 9 10 1 1 1, 2 1, 3 1, 2, 5 1, 3, 4 1, 2, 5, 6 1, 3, 4, 7 1, 3, 4, 7, 9 1, 2, 5, 6, 8 Route Reply 1, 3, 4, 7, 9, 10 1, 2, 5, 6, 8, 10 Remember link-wise disjoint paths between source and destination

17. Multipath Behavior  When primary route breaks, pick the best alternate route to forward data  When no alternate routes left, re-initiate route discovery

18. DSR – With Multipath (2nd Version) Route Request Message 1 2 3 5 4 7 6 8 9 10 1 1 1, 2 1, 3 1, 2, 5 1, 3, 4 1, 2, 5, 6 1, 3, 4, 7 1, 3, 4, 7, 9 1, 2, 5, 6, 8 Route Reply 1, 3, 4, 7, 9, 10 1, 2, 5, 6, 8, 10 Remember one link-wise disjoint path between each intermediate node and destination 3, 5, 6, 8, 10 4, 5, 6, 8, 10 7, 8, 109, 8, 10

19. Multipath Behavior  When a node detects a link is broken, the node replaces the unused path in the data packet with its alternate path  When alternate path breaks, the node transmits an error packet backward  Route discovery restarts only after all alternate paths between all intermediate nodes and destination have failed

20. DSR Properties  Multipath  On-demand  Loop-free (source routing)  Maintains connectivity

21. Performance Enhancements  Full use of route cache  Nodes operate in promiscuous mode to overhear route discoveries  Intermediary nodes must check for loop-free  Piggy-backing data on route discoveries

22. DSR Issues  Scalability issues:  Source routing means each packet carries the full path  MD5 - 16bytes per nodeID  10 nodes means 160bytes  11% of 1500 bytes packet  4X the size of TCP/IP header  IP - 4bytes per nodeID  10 nodes means 40bytes  3% of 1500 bytes packet  As big as the TCP/IP header

23. Basic AODV Route Discovery  When a route is needed, source floods a route request for the destination. S A B E C D RREQ (broadcast) From: U. of Cincinatti Marina & Das

24. Basic AODV Route Discovery  Reverse path is formed when a node hears a non- duplicate route request.  Each node forwards the request at most once (pure flooding). S A B E C D RREQ (broadcast) Reverse Path

25. Basic AODV Route Discovery  Reverse path is formed when a node hears a non- duplicate route request.  Each node forwards the request at most once (pure flooding). S A B E C D RREQ (broadcast) Reverse Path

26. Basic AODV Route Discovery  Observation: Duplicate RREQ copies completely ignored. Therefore, potentially useful alternate reverse path info lost. S A B E C D Reverse Path

27. Alternate Reverse Paths: A Naïve Approach  Form reverse paths using all duplicate RREQ copies.  Causes routing loops. S A B E C D Reverse Path

28. Alternate Reverse Paths: A Naïve Approach  Question: how to form alternate “loop-free” reverse paths using some duplicate RREQ copies? S A B E C D Reverse Path

29. Loop Freedom  Key idea: Impose ordering among nodes in every path.  Notion of upstream/downstream nodes.  General loop-freedom rule: Never form a route at a downstream node via an upstream node. d j i j is downstream to i w.r.t d

30. Multiple Loop-free Reverse Paths  Suppose RREQ from S includes highest seqno for itself. S A B E C D RREQ (broadcast) 0     

31. S A B E C D Reverse Path 0 1 1    Multiple Loop-free Reverse Paths

32. S A B E C D RREQ (broadcast) Reverse Path 0 1 1 2 2  Multiple Loop-free Reverse Paths

33. S A B E C D Reverse Path 0 1 1 2 2  Multiple Loop-free Reverse Paths

34. Alternate Paths  Keep multiple routes but always advertise only one hop count to others. Hop count of that path is the “advertised hop count”  Which one? Longest path at the time of first advertisement.  Or.. Keep only the disjoint alternate paths  The set of copy packets received at any node defines a set of disjoint paths  To ensure link-wise disjoint paths:  Exchange the first and last hops in packet  A different first and last hop ensures link-wise disjoint

35. AOMDV Properties  Multipath:  Can support local policies for choosing alternate paths  On-demand  Loop-free  Sequence numbers  Maintains connectivity

36. Final thoughts…  Unexplored trade-offs:  Reliability vs. energy consumption vs. benefit  Disjoint paths vs. braided paths  Bad interaction between TCP and on-demand adhoc protocols  The perennial problem of not being able to distinguish between congestion and broken link  When will TCP be using ECNs?  Crazy idea: use cell infrastructure to compute routing tables