1. Scalable Ad Hoc Routing the Case for Dynamic Addressing

2. <- DART valley, New Zealand Dynamic Address RouTing Jakob Eriksson, Michalis Faloutsos, Srikanth Krishnamurthy University of California, Riverside Dynamic Address RouTing Jakob Eriksson, Michalis Faloutsos, Srikanth Krishnamurthy University of California, Riverside DART

3. Routing in networks of millions of strangers. Be able to route packets even through handheld, low power, devices. Support wires, as well as directional and omnidirectional antennae. No assumption of infrastructure. No central ownership / administration. Plug-n-play operation, zero-configuration. Routing in networks of millions of strangers. Be able to route packets even through handheld, low power, devices. Support wires, as well as directional and omnidirectional antennae. No assumption of infrastructure. No central ownership / administration. Plug-n-play operation, zero-configuration. The Problem <- Poison DART frog

4. New/different approach to routing. Replace node address with two numbers: Node identifier - static. Routing address - dynamic. Dynamic routing address indicates current location in network topology. Proactive routing distributes routing info. Distributed lookup table maps identifier to current routing address. New/different approach to routing. Replace node address with two numbers: Node identifier - static. Routing address - dynamic. Dynamic routing address indicates current location in network topology. Proactive routing distributes routing info. Distributed lookup table maps identifier to current routing address. DART initial design

5. Why on Earth...? Rural networks Consumer owned networking Theater-wide military networks Internet 2.0? Civil disobedience Protecting civil liberties Free speech Ubiquitous and free connectivity? Networked society Who knew what Internet would become? Who knew what Internet would become? DIY wide-area networking Circumventing copyrights? Because we can! Developing countries <- motivation

6. Related Work DART Routing Address Allocation Node Lookup Simulations Conclusion Related Work DART Routing Address Allocation Node Lookup Simulations Conclusion Roadmap DART subway map, Dublin ->

7. Hierarchical routing (Kleinrock, Kamoun ‘77) Flat ad hoc routing protocols: AODV, DSR, DSDV. Numerous derivatives. Clustering based routing: Landmark, LANMAR, Clusterhead, Hierarchical State Routing, MMWN. Hybrids: ZRP, HARP, SHARP Georouting: LAR, DREAM, Grid, etc. Distributed Hashtables: Chord, Plaxton routing etc. Hierarchical routing (Kleinrock, Kamoun ‘77) Flat ad hoc routing protocols: AODV, DSR, DSDV. Numerous derivatives. Clustering based routing: Landmark, LANMAR, Clusterhead, Hierarchical State Routing, MMWN. Hybrids: ZRP, HARP, SHARP Georouting: LAR, DREAM, Grid, etc. Distributed Hashtables: Chord, Plaxton routing etc. Related Work Mobile IP? No, not quite related.

8. Routing Address - fixed length binary string. Address Prefix - sequence of most significant bits from routing address. Prefix Subgraph - The graph induced from a network graph by the set of nodes with a given prefix. Routing Address - fixed length binary string. Address Prefix - sequence of most significant bits from routing address. Prefix Subgraph - The graph induced from a network graph by the set of nodes with a given prefix. Some Terminology In DART, prefix subgraphs must be connected! 101 10x <- DART light rail

9. Address Space as Binary Tree Routing addresses form a virtual binary tree. All nodes within any given subtree are able to communicate using only nodes in that subtree. Routing addresses form a virtual binary tree. All nodes within any given subtree are able to communicate using only nodes in that subtree. Just prefixes, not actual nodes. Addresses that nodes can occupy.

10. Scalable routing through a virtual hierarchy. Each node keeps log N routing entries. Scalable routing through a virtual hierarchy. Each node keeps log N routing entries. The Routing Table

11. One unique address per node. All prefix subgraphs must be connected. Minimize communication overhead. No centralized resources/infrastructure. Minimize address size (in bits). One unique address per node. All prefix subgraphs must be connected. Minimize communication overhead. No centralized resources/infrastructure. Minimize address size (in bits). The Addressing Problem IMPORTANT! <- DART automobile

12. Basic Solution example When a node joins, it picks an address that shares a prefix one of its neighbors. Neighbors’ routing tables show valid address ranges that are still unoccupied. When a node joins, it picks an address that shares a prefix one of its neighbors. Neighbors’ routing tables show valid address ranges that are still unoccupied.

13. Node Lookup Table Efficient distributed hashtable (DHT) that maps Node ID -> Routing Address. Similar to overlay DHT research, but uses existing routing layer state for efficiency. Nodes periodically send their current address to one other node, their lookup peer. Upon connection establishment, the routing address of the destination retrieved from the destination’s lookup peer. Efficient distributed hashtable (DHT) that maps Node ID -> Routing Address. Similar to overlay DHT research, but uses existing routing layer state for efficiency. Nodes periodically send their current address to one other node, their lookup peer. Upon connection establishment, the routing address of the destination retrieved from the destination’s lookup peer.

14. Lookup Table Basics Node 3 (011) “Node 3 has address 001” Node 7 (111) “Node 7 has address 000” ?? Every ID->Address mapping is stored at the node whose address most closely matches the ID!

15. Performed using two simulators: Home-grown for visual feedback, speed and scalability. ns-2 for comparisons and accuracy. Wireless nodes, with omnidirectional antennae. Performed using two simulators: Home-grown for visual feedback, speed and scalability. ns-2 for comparisons and accuracy. Wireless nodes, with omnidirectional antennae. Simulations DART personal aircraft ->

16. Extremely small average routing table size < 2*log N. About 15 entries for 4000 nodes! Extremely small average routing table size < 2*log N. About 15 entries for 4000 nodes! Routing Table Size

17. Low average path stretch, 30-35%, so route aggregation is not hurting us much. Experimental Results

18. Ns-2 doesn’t scale to large wireless simulations. Simulated 400-node networks. Varied connection establishment frequency (CEF). Arguably, CEF increases in larger networks. Also, CEF depends on traffic patterns. Ns-2 doesn’t scale to large wireless simulations. Simulated 400-node networks. Varied connection establishment frequency (CEF). Arguably, CEF increases in larger networks. Also, CEF depends on traffic patterns. Simulating large networks

19. Overhead vs. CEF DART overhead is not affected by CEF. Reactive protocols suffer when CEF increases. DART overhead is not affected by CEF. Reactive protocols suffer when CEF increases. Updates Updates & Requests

20. Throughput vs. CEF DART reliably outperforms AODV/DSR when connection establishment frequency > 3. CEF == 3 means one connection/node every 2 mins. DART reliably outperforms AODV/DSR when connection establishment frequency > 3. CEF == 3 means one connection/node every 2 mins.

21. Dynamic Addressing represents a novel and promising approach to scalable ad hoc routing. DART is on its way to become a scalable alternative to current ad hoc routing protocol. Simulation results indicate significant performance improvement even for 400-node networks. Dynamic Addressing represents a novel and promising approach to scalable ad hoc routing. DART is on its way to become a scalable alternative to current ad hoc routing protocol. Simulation results indicate significant performance improvement even for 400-node networks. Conclusion <- DART valley, New Zealand

22. The End <- DART simulator dart.cs.ucr.edu