1. A Membership Management Protocol for Mobile P2P Networks Mohamed Karim SBAI, Emna SALHI, Chadi BARAKAT

2. Mobile Ad hoc Networks  Spontaneous multi-hop wireless networks  end-to-end communication  ad hoc routing protocols   Without any established infrastructure   Nodes play symmetric roles  No dedicated nodes.   Using wireless channel   Limited and shared resources   Mobility  Network splits 

3. P2P Networks  Peer-to-peer services (as known in the Internet)  Without dedicated devices (servers)   Peers play symmetric roles  Both clients and servers.   Can use fixed servers to track the members of the overlay   The mechanism are not adapted to mobile constrained environments 

4. Membership Management Protocol for mobile P2P networks  Objective: Maintaining an up-to-date list of the peers interested in the P2P service.  Challenges: - Minimum cost on the underlying network. - Ensuring the continuity of the service. - Having a good level of the freshness of information.

5. A membership management protocol for P2P services run over MANET ?  Client / Server   Flooding-based method   Multicast-based method   P2P   Adaptive and optimal P2P method ?

6. Membership Management Protocol  Our solution: A fully distributed protocol for constructing and maintaining minimum spanning trees of interested peers.  robust  adaptive  network friendly  decentralized  Algorithms: 1. Joining the membership tree 2. Leaving the membership tree 3. Adapting the membership tree to mobility of nodes 4. Network split awareness

7. Joining the membership tree  Looking for the nearest peer  a controlled-scope flooding method  Connecting to the nearest peer and getting the current tree from it  Dissemination of the new arrival information on the tree  Changing some connections of the tree considering the cut property of a minimum spanning tree.

8. Adapting the tree to mobility of nodes  Two peers that are neighbors in the tree can get closer  the tree is still optimal.  Two peers that are not neighbors in the spanning tree get farther from each other  the cost of the tree does not change and no better decision can be made.  Two peers that are neighbors in the spanning tree get farther from each other.  The cost of the tree increases  there might exist a better tree. CASE 1  Two peers that are not neighbors in the spanning tree get closer to each other  It might be another tree with smaller weight. CASE 2

9. Adapting the tree to mobility of nodes  CASE I = CASE 2 If one of the peers get nearer to another peer in the tree. Else, no optimization can be made.  CASE 2 : Using the cycle property of a minimum spanning tree to elect the logical link to cut.

10. Leaving the membership tree  The child of the leaving peer having the highest identifier connects to its parent and becomes the parent for the remaining children.  A new spanning tree  The optimal is reached by having the peers apply the normal approaching adaptation procedure.

11. Network split awareness  Tagging network nodes that are not interested in the same service.  Tracks continuously the appearance of non tagged nodes in its neighborhood.  A new node not tagged and not belonging to the same membership tree is a good candidate to be asked whether it belongs to the same service but comes from another cluster.  Executing a join procedure in case the node is a peer.

12. Packet format

13. Performance evaluation  Performance metrics:  Real cost: number of hops message  Cost corrected by freshness of information  NS-2 Simulations scenario :  50 nodes / Random way point (2ms, 30s) / OLSR routing protocol  exponentiel distribution of ON and OFF times of peers

14. Performance evaluation Client/server method

15. Performance evaluation

21. Thank You mksbai@sophia.inria.fr