Using Directionality in Mobile Routing BowNan Cheng MIT

Using Directionality in Mobile Routing Bow-Nan Cheng (MIT LL) Murat Yuksel (Univ Nevada - Reno) Shivkumar Kalyanaraman (IBM IRL) (Work done at Rensselaer Polytechnic Institute) 1

Motivation Infrastructure / Wireless Mesh Networks • Characteristics: Fixed, unlimited energy, virtually unlimited processing power • Dynamism – Link Quality • Optimize – High throughput, low latency, balanced load Scalability Layer 3: Network Layer Mobile Adhoc Networks (MANET) • Characteristics: Mobile, limited energy • Dynamism – Node mobility + Link Quality • Optimize – Reachability Sensor Networks Main Issue: Scalability Introduction MORRP Key Concepts • Characteristics: Data-Centric, extreme limited energy • Dynamism – Node State/Status (on/off) • Optimize – Power consumption Simulation Results Conclusion 2

Scaling Networks: Trends in Layer 3 Flood-based Mobile Ad hoc / Fixed Wireless Networks DSR, AODV, TORA, DSDV Partial Flood: OLSR, HSLS Peer to Peer / Gnutella Overlay Networks Wired Networks Introduction OSPF, IEGRP, RIP Hierarchy/Structured Unstructured/Flat Scalable LGF, VRR, GPSR+GLS Hierarchical Routing, WSR (Mobicom 07) ORRP (ICNP 06) Kazaa, DHT Approaches: CHORD, CAN Bubble. Storm (Sigcomm 07) LMS (PODC 05) OSPF Areas MORRP Key Concepts Simulation Results Conclusion 3

Trends: Directional Communications Directional/Directive Antennas B’ B’ B A A’ Hybrid FSO / RF MANETS B D’ A D C C’ Omni-directional A’ D’ D C C’ Directional • Directional Antennas – Capacity Benefits § Theoretical Capacity Improvements - factor of 4 p 2/sqrt(ab) where a and b are the spreads of the sending and receiving transceiver ~ 50 x capacity with 8 Interfaces (Yi et al. , 2005) § Sector Antennas in Cell Base Stations – Even only 3 sectors increases capacity by 1. 714 (Rappaport, 2006) Introduction MORRP Key Concepts • Current RF-based Ad Hoc Networks: § omni-directional RF antennas § High-power – typically the most power consuming parts of laptops § Low bandwidth § Error-prone, high losses § Free Space Optics: § High bandwidth § Low Power § Dense Spatial Reuse § License-free band of operation Simulation Results Conclusion 4

ORRP Big Picture Orthogonal Rendezvous Routing Protocol ORRP Primitive 1: Local sense of direction leads to ability to forward packets in opposite directions ORRP 180 o S T 2: Forwarding along Orthogonal lines has a high chance of intersection in area Introduction A 98% • High reach (98%), O(N 3/2) State complexity, Low path stretch (~1. 2), Up to 69% high goodput, unstructured • BUT. . What happens with mobility? Wireless Mesh Networks Mobile Ad-Hoc Networks B Increasing Mobility 65% 55% 42% Overlay Networks 5

Mobile-ORRP (MORRP) Introduction What can we do? A a R B Introduction MORRP Key Concepts • Replace intersection point with intersection region. • Shift directions of send based on local movement information • Route packets probabilistically rather than based on rigid nexthop paths. (No need for route maintenance!) • Solution: a NEW kind of routing table: Directional Routing Table (DRT) Simulation Results Conclusion 6

MORRP Basic Example C R: Near Field DRT Region of Influence B S A R F Original Path S O Q N R P S: Near Field DRT Region of Influence G R’ E Original Direction (a 1) Original Path New Direction (a 2) M D L D’ I H K D: Near Field DRT Region of Influence J 1. Proactive Element – Generates Rendezvous to Dest Paths 2. Reactive Element – Generates Source to Rendezvous Paths Introduction MORRP Key Concepts Simulation Results Conclusion 7

The Directional Routing Table Use Decaying Bloom Filter (DBF) Routing Tables viewed from Node A 4 Z 3 D A 2 1 B Routing Table C RT w/ Beam ID Directional RT (DRT) Dest ID Next Hop Beam ID Dest IDs (% of Certainty) Beam ID B C D : Z B B Z : Z 1 1 3 : 3 B(90%), C(30%). Z(90%), D(40%). 1 2 3 4 ID ID ID set of IDs Set of IDs set of IDs • Soft State – Traditional routing tables have a hard timeout for routing entries. Soft State decreases the level of certainty with time. • Uncertainty with Distance – Nodes closer to a source will have increasingly more information about the location of the source than nodes farther away • Uncertainty with Time – As time goes on, without updates, one will have lesser amount of information about the location of a node • Uncertainty with Mobility – Neighbors can potentially be “covered” by different interfaces based on mobility speed and direction Introduction MORRP Key Concepts Simulation Results Conclusion 8

DRT Intra-node Decay Time Decay with Mobility Spread Decay with Mobility a q 2 7 x x q 3 q 1 8 As node moves in direction +x, the certainty of being able to reach nodes covered by region 8 should decay faster than of region 7 depending on speed. This information is DROPPED. Introduction q 2 > q 1 > q 3 MORRP Key Concepts a As node moves in direction +x, the certainty of being able to reach nodes covered by region 2 should be SPREAD to region 1 and 3 faster than the opposite direction. The information about a node in region 2 should be SPREAD to regions 1 and 3. Simulation Results Conclusion 9

MORRP Fields of Operation N N N N S R N N N • Near Field Operation § Uses “Near Field DRT” to match for nodes 2 -3 hops away • Far Field Operation § RREQ/RREP much like ORRP except nodes along path store info in “Far. Field DRT” Introduction MORRP Key Concepts Simulation Results N N D N N Conclusion 10

Performance Evaluation of MORRP • Metrics Evaluated § Reachability – Percentage of nodes reachable by each node in network (Hypothesis: high reachability) § Delivery Success – Percentage of packets successfully delivered network-wide § Scalability – The total state control packets flooding the network (Hypothesis: higher than ORRP but lower than current protocols out there) § Average Path Length § End to End Delay (Latency) § Aggregate Network Goodput • Scenarios Evaluated (NS 2) § Evaluation of metrics vs. AODV (reactive), OLSR (proactive), GPSR with GLS (position-based), and ORRP under various node velocities, densities, topology-sizes, transmission rates. § Evaluation of metrics vs. AODV and OLSR modified to support beamswitched directional antennas. Introduction MORRP Key Concepts Simulation Results Conclusion 11

MORRP: Aggregate Goodput Results • Aggregate Network Goodput vs. Traditional Routing Protocols § MORRP achieves from 10 -14 X the goodput of AODV, OLSR, and GPSR w/ GLS with an omni-directional antenna § Gains come from the move toward directional antennas (more efficient medium usage) • Aggregate Network Goodput vs. AODV and OLSR modified with directional antennas § MORRP achieves about 15 -20% increase in goodput vs. OLSR with multiple directional antennas § Gains come from using directionality more efficiently Introduction MORRP Key Concepts Simulation Results Conclusion 12

MORRP: Simulations Summary • MORRP achieves high reachability (93% in mid-sized, 1300 x 1300 m 2 and 87% in large-sized, 2000 x 2000 m 2 topologies) with high mobility (30 m/s). • With sparser and larger networks, MORRP performs fairly poorly (83% reach) suggesting additional research into proper DRT tuning is required. • In lightly loaded networks, MORRP end-to-end latency is double of OLSR and about 7 x smaller than AODV and 40 x less than GPSR w/ GLS • MORRP scales well by minimizing control packets sent • MORRP yields over 10 -14 X the aggregate network throughput compared to traditional routing protocols with one omnidirectional interface gains from using directional interfaces • MORRP yields over 15 -20% the aggregate network goodput compared to traditional routing protocols modified with 8 directional interfaces gains from using directionality constructively Introduction MORRP Key Concepts Simulation Results Conclusion 13

MORRP: Key Contributions • The Directional Routing Table § A replacement for traditional routing tables that routes based on probabilistic hints § Gives a basic building block for using directionality to overcome issues with high mobility in MANET and DTNs • Using directionality in layer 3 to solve the issues caused by high mobility in MANETs • MORRP achieves high reachability (87% - 93%) in high mobility (30 m/s) • MORRP scales well by minimizing control packets sent • MORRP shows that high reach can be achieved in probabilistic routing without the need to frequently disseminate node position information. • MORRP yields high aggregate network goodput with the gains coming not only from utilizing directional antennas, but utilizing the concept of directionality itself. • MORRP is scalable and routes successfully with more relaxed requirements (No need for coordinate space embedding) Introduction MORRP Key Concepts Simulation Results Conclusion 14

Thank You! • Questions and Comments? • Papers / Posters / Slides / NS 2 Code (MORRP, OLSR + AODV with Beam switched directional antennas) [ http: //networks. ecse. rpi. edu/~bownan ] • bownan@gmail. com Introduction MORRP Key Concepts Simulation Results Conclusion 15