A Probabilistic Routing Protocol for Mobile Ad Hoc Networks Abdallah Jabbour • James Psota • Alexey Radul {ajabbour, psota, axch}@mit.edu Jabbour, Psota, Radul 6.829 Final Project
Outline Related Routing Protocols Shortcomings of related protocols DSDV, DSR, AODV Probabilistic routing protocols Shortcomings of related protocols Protocol description Simulation environment Measures of evaluation Simulation results Conclusions and future work Jabbour, Psota, Radul 6.829 Final Project
Related Routing Protocols Destination-Sequenced Distance Vector (DSDV) Hop-by-hop distance vector protocol Routes tagged with sequence numbers Proactive Dynamic Source Routing (DSR) On-demand source routing Floods route requests Maintains routes by link breakage notification Ad Hoc On-Demand Distance Vector (AODV) Borrows sequence numbers from DSDV and the Route Discovery mechanism from DSR Uses RREQ, RREP, RREP ACK, RERR and HELLO packets Jabbour, Psota, Radul 6.829 Final Project
Probabilistic Routing Protocols Routing table entries have probability values corresponding to each destination-neighbor pair Control packets (“ants”) sent randomly Data forwarded deterministically along path with best metric (number of hops) Examples Ant-Based Control (ABC) AntNet Ant-Colony-Based Routing Algorithm (ARA) Jabbour, Psota, Radul 6.829 Final Project
Drawbacks and Limitations of Above Protocols Routing packets hinder performance Decrease available bandwidth Increase transmission latency High recovery latency due to static routes DSDV, DSR, AODV Probabilistic protocols incorrectly assume symmetric traffic Above protocols use shortest hop routes Tend to pick routes with less capacity than optimal ones Tend to use marginal links Jabbour, Psota, Radul 6.829 Final Project
Questions that need answers Is it possible to minimize routing packets? - Especially those interfering with traffic How can nodes cooperate with little or no control traffic? Can one make forwarding decisions based on a better measure of network state? How can one better cope with link outages? Which is better: random routing or deterministic routing? our goals are to increase performance by… Jabbour, Psota, Radul 6.829 Final Project
The answers! Control packets are minimized by prepending protocol-level headers onto all data packets Both when originating and forwarding a packet Nodes cooperate by promiscuously listening to all traffic, using protocol headers to update their state Routing decisions are based on link loss ratios ETX used instead of minimum hop count Probabilistic routing is made modular - choice of metric - choice of metric-to-probability mapping - choice of routing strategy (random or deterministic) Jabbour, Psota, Radul 6.829 Final Project
Protocol Header Contents Each originated or forwarded packet contains the following protocol-level header: Jabbour, Psota, Radul 6.829 Final Project
Node State Nodes maintain the following state Dynamically-updated set of neighbors Loss ratios to and from each neighbor Routing state Metric values for each destination and each destination-neighbor pair Probability of forwarding to a certain neighbor in order to reach a desired destination Requests for and fulfillments thereof information about destinations Jabbour, Psota, Radul 6.829 Final Project
State Update Nodes update state Probability distribution updates Upon sending Upon receiving Periodically Refresh stale state and, if needed, alert neighbors that you’re still alive Probability distribution updates Probability distribution and metric values updated along with other node state Values evolve in response to changes in link quality and to nodes entering and leaving the system Jabbour, Psota, Radul 6.829 Final Project
Probabilistic Routing routing table p1 = 0.1 dest p1 p2 p3 d 0.1 0.4 0.5 n1 n2 n3 d 0.1 0.4 0.5 dest p1 p2 p3 d 0.1 0.4 0.5 s n2 d p1 = 0.4 p3 = 0.5 n3 Route is not fixed, so packets can still reach destination immediately upon link breakage Jabbour, Psota, Radul 6.829 Final Project
Probabilistic Routing routing table p1 = 0.3 n1 n2 n3 d 0.1 0.3 0.4 0.0 0.5 0.7 x s n2 d p1 = 0.4 x x x link breaks! p3 = 0.7 n3 Update forwarding probability upon link breakage Jabbour, Psota, Radul 6.829 Final Project
Probabilistic Routing Strategies Random: node forwards probabilistically to neighbor ni with probability pi Deterministic: node forwards ALL data packets along path with highest pi Our flexible infrastructure allowed simulation of both First to compare random to deterministic routing Jabbour, Psota, Radul 6.829 Final Project
Simulation Environment ns-2 with Monarch mobility extensions Compared the new protocol to DSDV, DSR and AODV 50 mobile nodes in a 1500m x 300m area Random waypoint movement model 900s simulation time Used UDP(CBR) sources TCP’s inconvenience: conforming load We investigated different… Pause times Node speeds Connection patterns Packet sizes Jabbour, Psota, Radul 6.829 Final Project
Measures of Evaluation Packet delivery ratio/ goodput Packet delivery latency Routing packet overhead Total bytes of overhead Path length optimality Route acquisition latency Jabbour, Psota, Radul 6.829 Final Project
Simulation Results Jabbour, Psota, Radul 6.829 Final Project
Conclusions and Future Work Jabbour, Psota, Radul 6.829 Final Project