1. DHT-based unicast for mobile ad hoc networks Thomas Zahn, Jochen Schiller Institute of Computer Science Freie Universitat Berlin 報告 : 羅世豪

2. Outline 一. Introduction  MADPastry 三. MADPastry’s Unicast  Simulation Results  Conclusion

3. Introduction (1/4)  Both MANETs and peer-to-peer (P2P) overlay networks share a number of key characteristics.  the lack of a central infrastructure  highly dynamic network topologies  the need for self-organization

4. Introduction (2/4)  Distributed Hash Tables (DHTs) have been proposed for large-scale network applications.  Route a packet based on a key (rather than a fixed destination address) to the (unknown) node in the network that is currently responsible for the given key within a bounded number of hops.  This overlay routing is also referred to as indirect routing.

5. Introduction (3/4)  DHTs do not closely concern themselves with the physical (routing) aspects of the underlying network  designed to form overlay networks in Internet-based networks where physical routing is practically taken for granted.

6. Introduction (4/4)  In MANETs, DHT can provide  indirect routing  direct physical routing (unicasting)  MADPastry  a DHT substrate explicitly designed for MANETs  DHT-based unicast can achieve  better packet delivery rates  lower network traffic

7. MADPastry (1/8)  MADPastry combines AODV ad hoc routing and Pastry overlay routing at the network layer  provide an efficient primitive for key-based routing in MANETs.

8. MADPastry (2/8)  Each node in a MADPastry network assigns itself a unique overlay id  defines its logical position on the virtual overlay id ring.  A message's packet header contains a message key.  MADPastry routes the message to that node in the network that is currently responsible for the message key  the node whose overlay id is currently the numerically closest to the message key among all MADPastry nodes in the network.  To avoid message broadcasts (e.g. for route discovery)  MADPastry considers physical locality in the construction of its routing tables.

9. MADPastry (3/8) - Clusters  Random Landmark  No fixed landmark nodes, landmark keys instead  These keys divide the logical overlay id space into equal sections  0800..00, 1800..00,......., F800..00  Node currently closest to a landmark key  Become temporary landmark node  Periodically issue beacon messages  Nodes overhear these beacon messages and periodically determine the physically closest temporary landmark node.

10. MADPastry (4/8) - Clusters  Node associates itself with closest temporary landmark  assumes same overlay ID prefix  If need be, a node assigns itself a new overlay id sharing the same prefix with the new closest temporary landmark node.  physically close nodes forming overlay clusters with common id prefixes  Physically close nodes are also likely to be close in the overlay

11. MADPastry (5/8)- Spatial Topology Spatial Topology

12. MADPastry (6/8) – Routing Tables  MADPastry maintains three different routing tables:  AODV-style routing table for physical routes from a node to specific target nodes  stripped down Pastry routing table that only contain landmark keys  standard Pastry leaf set for indirect routing

13. MADPastry (7/8) – Routing Tables

14. MADPastry (8/8) - Routing  When a node wants to send a packet to a specific key 1.Consults its Pastry routing table and/or leaf set to determine the closest prefix match, as stipulated by standard Pastry. 2.Consults its AODV routing table for the physical route to execute this overlay hop. 3.Intermediate nodes on the physical path of an overlay hop consult their AODV table for the corresponding next physical hop. 4.When a packet reaches the destination of an overlay hop, that node again consults its Pastry routing table and/or leaf set to determine the next overlay hop.  This process continues until the packet reaches the eventual target node that is responsible for the packet key ( whose overlay id is the numerically closest to the packet key).

15. MADPastry’s Unicast (1/7) – Address Publication  Each nodes has exactly one temporary address server  Address server stores its client's current overlay ID  Node A hashes its node ID  address server key (ASK).  Node A publishes its current overlay ID towards ASK  Node currently responsible for node A's hash key becomes node A's address server

16. MADPastry’s Unicast (2/7) – Address Publication ASK:h(17) →B7A9CC

17. MADPastry’s Unicast (3/7) – Address Resolution  Node A wants to communicate with node B  Node A does not know node B's current overlay ID  Node A hashes node B's net ID to get ASK  Node A sends request towards ASK  Node B's address server replies with node B's current overlay ID

18. MADPastry’s Unicast (4/7)– Address Resolution ASK: h(17) →B7A9CC

19. MADPastry’s Unicast (5/7) – Unicast  Node A uses overlay ID from reply to send message to node B  MADPastry delivers message using indirect routing

20. MADPastry’s Unicast (6/7) – Unicast

21. MADPastry’s Unicast (7/7)  Shortest Routing Paths  A direct, straight path from the source node to the destination node  When using AODV and it doesn’t know the path to that destination, it needs to engage in a costly route discovery process  MADPastry’s Routing Paths  Probably travel multiple overlay hops  The entries in its routing table are likely up-to-date and valid  Usually won’t have to discover a route because it can use at each overlay routing step any existing route that would bring the packet numerically closer to its key

22. Simulation Results (1/6)  Compare MADPastry's unicast against a popular ad hoc routing protocol  AODV (reactive)  OLSR (proactive)  Simulations in ns2  1 random request every 10s per node

23. Simulation Results (2/6)  Constant node velocity:1.4 m/s  Varying network sizes (50,100,150,200,250)

24. Simulation Results (3/6)

25. Simulation Results (4/6)

26. Simulation Results (5/6)  Constant network size:250 nodes  Varying node velocities (0.1,1.4, 2.5,5.0 m/s)

27. Simulation Results (6/6)

28. Conclusion (2/2)  MADPastry already present in the MANET to provide key-based, indirect routing  MADPastry can also provide point-to-point unicasting  No need to maintain ad hoc routing protocol in parallel for DHT applications that use MADPastry handle their point-to-point routing

29. Conclusion (2/2)  MADPastry's unicast can outperform popular reactive and proactive ad hoc routing protocols  In MANETs it can be advantageous to travel numerous short up-to-date routes instead of one long direct route