Volcano Routing Scheme Routing in a Highly Dynamic Environment Yashar Ganjali Stanford University Joint work with: Nick McKeown SECON 2005, Santa Clara, CA, Sep. 27,
September 2005 Volcano Routing Scheme 2 Outline Routing in MANETs Slowly changing topology Highly changing topology Volcano Routing Scheme Single Flow Multiple Flows Evaluation Mathematical Results Simulations
September 2005 Volcano Routing Scheme 3 Routing in Data Networks Routing in data networks Phase 1: Route discovery Proactive Reactive or on-demand Phase 2: Packet forwarding Routing overhead is reduced Discovery happens very infrequently s d
September 2005 Volcano Routing Scheme 4 Routing in MANETs Changes in topology Node movements Wireless link issues Route changes more frequent Temporary partitioning in network Increased overhead of route discovery phase Accelerate/defer the route discovery process Use flooding to find routes as quickly as possible Buffer when partitioned
September 2005 Volcano Routing Scheme 5 Highly Dynamic Topology What if topology changes constantly? Quickly moving nodes Highly dynamic environment Adversarial model Route discovery failure two-phase routing doesn’t work
September 2005 Volcano Routing Scheme 6 One-Phase Routing Eliminate explicit route discovery Assign a function to nodes that determines the direction of packets Physical location of nodes: Some variations of geographical routing Number of packets buffered in a node: Volcano Routing Scheme (VRS)
September 2005 Volcano Routing Scheme 7 Outline Routing in MANETs Slowly changing topology Highly changing topology Volcano Routing Scheme Single Flow Multiple Flows Evaluation Mathematical Results Simulations
September 2005 Volcano Routing Scheme 8 Volcano Routing Scheme (VRS) Lava flows towards the sea (low altitude) Local balancing of load Obstacles do not stop lava No explicit route discovery Reordering layers doesn’t disrupt the flow
September 2005 Volcano Routing Scheme 9 At the beginning of each time slot: Packets are generated at the source. During the time slot: Each link (v,w) for which P(v) – P(w) > transfers one packet from v to w. is called transfer threshold. At the end of the time slot: Packets which arrive at destination are removed.
September 2005 Volcano Routing Scheme 10 Simple Example Time slot 1 Packet generated Time slot 2 Packet generated Two transfered One received Time slot 3 Packet generated Time slot 4 Packet generated One transfered One received … sd m
September 2005 Volcano Routing Scheme 11 Volcano Routing Scheme
September 2005 Volcano Routing Scheme 12 Pros and Cons Advantages No explicit route discovery Completely distributed Low complexity Minimal amount of control traffic Suitable for highly dynamic environments System is proved to be stable Path taken by packets is near optimal Limitations Requires continuous stream of packets from source to destination Packet reordering might happen
September 2005 Volcano Routing Scheme 13 Multi-Flow VRS Time-Division VRS Divide time equally among K flows Maximum-Pressure VRS For a link (v,w) serve the flow i which has the maximum amount of pressure P i (v)- P i (w)
September 2005 Volcano Routing Scheme 14 Multi-Flow VRS
September 2005 Volcano Routing Scheme 15 Outline Routing in MANETs Slowly changing topology Highly changing topology Volcano Routing Scheme Single Flow Multiple Flows Evaluation Mathematical Results Simulations
September 2005 Volcano Routing Scheme 16 Evaluation Method Metrics Stability (packet loss ratio) Queue size distribution Routing path length Factors Connectivity (communication range, number of nodes, …) Number and amount of flows Mobility process Transfer threshold
September 2005 Volcano Routing Scheme 17 Stability Strict Stability: total number of packets in the network is bounded. F-Min-Provisioned: capacity of minimum cut is at least F. Theorem. If the source injects at most F packets the system remains strictly stable if the network is F-min- provisioned. s d
September 2005 Volcano Routing Scheme 18 Packet Loss vs. Flow Demand 100 nodes distributed uniformly in a 1x1 square CR = 0.26 Velocity ~ [ ] = 2 Average number of neighbors = 20 Stability independent of buffer size
September 2005 Volcano Routing Scheme 19 Packet Loss: TD-VRS vs. MP-VRS
September 2005 Volcano Routing Scheme 20 Packet Loss: Communication Range Average No. of Neighbors = Flow Demand
September 2005 Volcano Routing Scheme 21 Packet Loss: Mobility Process No difference between random walk and waypoint model Stability independent of velocity Extremely low velocity can cause instability
September 2005 Volcano Routing Scheme 22 Queue Size Distribution
September 2005 Volcano Routing Scheme 23 Near-Optimal Paths In a fixed topology packets take shortest paths. If flow rate is D- we can choose such that Almost surely all packets take the first D shortest paths. Trade-ff between Number of outstanding packets Routing path length
September 2005 Volcano Routing Scheme 24 Path Length vs. Delta
September 2005 Volcano Routing Scheme 25 Summary Introduced Volcano Routing Scheme Distributed, fast, low complexity, … Need stream of packets Variations of VRS: Time Division, Maximum Pressure Stable under admissible traffic Short queuing delay Routing path near optimal
September 2005 Volcano Routing Scheme 26 Thank You! Questions?
September 2005 Volcano Routing Scheme 27 Extra Slides
September 2005 Volcano Routing Scheme 28 Generalizing to More Flows Flow 1 Source: node 1 Destination: node 4 Flow 2 Source: node 4 Destination: node 1
September 2005 Volcano Routing Scheme 29 Packet Loss: Flow Demand
September 2005 Volcano Routing Scheme 30 Packet Loss: Number of Nodes
September 2005 Volcano Routing Scheme 31 Loss vs. Velocity
September 2005 Volcano Routing Scheme 32 Packet Loss vs. Node Velocity
September 2005 Volcano Routing Scheme 33 Queue Size Distribution: Delta
September 2005 Volcano Routing Scheme 34 Queue Size Distribution