Weak State Routing for Large Scale Dynamic Networks Utku Gnay Acer, - PowerPoint PPT Presentation

Weak State Routing for Large Scale Dynamic Networks Utku Günay Acer, Shivkumar Kalyanaraman, Alhussein A. Abouzeid Rensselaer Polytechnic Institute Department of Electrical, Computer & Systems Engineering

q Which area should we NOT be working on in MOBICOM anymore? Ans: Routing ! q - Victor Bahl, Mobicom 2007 panel

Routing in Large-scale Dynamic Networks q Routing table entries: “ state ” = indirections from persistent names (ID) to locators Node Mobility q Due to dynamism, such indirections break q Problematic on two dimensions þ Dynamism/mobility => frequent update of state þ Dynamism + large scale => very high overhead, hard to maintain structure Number of Nodes q We propose constructing routing table indirections using probabilistic and more stable state: WEAK STATE

A new class of State q Strong State q Weak State þ Deterministic þ Probabilistic hints þ Requires control þ Updated locally traffic to refresh þ Rapidly invalidated þ Exhibits in dynamic persistence environments

Hard, Soft and Weak State REMOVE UPDATE INSTALL a b STATE STATE STATE STATE A B A B Time elapsed since Confidence in state state installed/ information Weak State Hard State Soft State refreshed Weak State is natural generalization of soft state

Random Directional Walks q Both used to announce location information ( “ put ” ) and forward packets ( “ get ” )

Outline q Our Weak State Realization q Disseminating Information and Forwarding Packets q Simulation Results q Discussion & Conclusion

An Instance of Weak State SetofIDs GeoRegion {a,b,c,d,e,f} Probabilistic Probabilistic in terms of in terms of scope membership q The uncertainty in the mappings is captured by locally weakening/decaying the state q Other realizations are possible þ Prophet, EDBF etc…

Example: Weak State q Consider a node a q Where is node a? þ (i): It is in region AB with probability � 1 þ (ii) It is in region B with probability � 2 ( � 1 · � 2 )

How to “ Weaken ” State? Larger Geo-Region 2 x Aggregation 128.113.50. 128.113.62. x x 1 x 128.113. θ 2 θ 1 x n SetofIDs -> GeoRegion

Aggregation: setofIDs q setofIDs: We use a Bloom filter, denoted by B . m 1 m 2 … . … . k k … . 0 1 0 0 0 1 0 1 0 1 1 1 0 1 u h j (m 1 ) h i (m 2 ) h i (m 1 ) h j (m 2 )

Decaying/Weakening the setofIDs q Randomly reset 1 ’ s to 0. Same as EDBF [Kumar et al. Infocom ’ 05] … . … . … . … . … . 0 1 0 0 0 0 1 0 0 0 0 1 0 0 0 0 1 0 0 0 0 1 0 0 0 1 1 1 1 1 0 0 0 0 0 0 1 1 0 0 0 0 0 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 0 1 0 1 0 1 0 1 0 h j (m) h 1 (m) h i (m) h k (m) q Let � (m)= � i=1 m B(h i (m)) q Large � (m) ! fresh mapping q � (m)/k is a rough measure of P{x m 2 A}

Weakening State (Contd) u u B ≤ B ≥ 2 2 x n x n x n setofIDs small, time passes: Either setofIDs large OR Decay GeoRegion GeoRegion Large => Decay SetofIDs

Random Directional Walks q Both used to announce location information ( “ put ” ) and forward packets ( “ get ” )

Dissemination/Proactive Phase: (put) q When a node receives a location C announcement, it þ creates a ID-to- location mapping B þ aggregates this mapping with A previously created mappings if possible

Forwarding Packets (get) A S Confidence Confidence Confidence Confidence Confidence Confidence Confidence Confidence 0.71 0.71 0.71 0.71 0.71 0.71 0.71 0.71 B C WSR involves unstructured, flat, but Confidence Confidence Confidence Confidence Confidence Confidence 0.84 0.84 0.84 0.84 0.84 0.84 scalable routing ; no flooding ! E 1.0 1.0 D Confidence Confidence Confidence Confidence 1.0 1.0 1.0 1.0 Forwarding decision: similar to longest-prefix-match. “ strongest semantics match ” to decide how to bias the random walk.

Simulation Objectives q How does WSR perform? þ Large-scale þ High Mobility q Comparisons: þ DSR: works well for small scale, high mobility þ GLS+GPSR: works well for large scale, low mobility q Short answer: 98%+ packet delivery despite large scale AND high mobility. q Tradeoffs: longer path lengths, � (N 3/2 ) state overhead

Simulation Setup q Benchmarks þ DSR and GLS-GPSR q Random waypoint mobility model with v min =5 m/s and v max =10 m/s þ WSR is robust against dynamism (please see the paper for details) q Performance Metrics þ Packet delivery ratio þ Control packet overhead þ Number of states maintained þ Normalized path length þ End-to-end Delay

Packet Delivery Ratio 1.2 WSR GLS-GPSR 1 DSR 0.8 Packet Devivery Ratio GLS breaks down due to DSR only delivers a WSR achieves high overheads 0.6 small fraction of delivery ratio packets 0.4 0.2 0 -0.2 400 500 600 700 800 900 1000 Number of Nodes

Control Packet Overhead 7000 WSR Total Overhead per Second (Number of Packets) GLS-GPSR 6000 5000 4000 Maintaining structure requires superlinearly 3000 increasing overhead in GLS 2000 1000 0 400 500 600 700 800 900 1000 Number of Nodes

Number of States Maintained 4 x 10 4 The total state stored in Total Number of Mappings/Database Entries Maintained WSR the network increases as GLS-GPSR 3.5 � (N 3/2 ) instead of � (NlogN) 3 2.5 2 1.5 1 0.5 0 400 500 600 700 800 900 1000 Number of Nodes

Per-Node State Dynamics 45 40 35 Number of States Maintained 30 State generation rate matches state 25 removal rate. 20 15 10 5 0 0 100 200 300 400 500 600 700 800 900 1000 time (seconds)

Distribution of Per-Node State in the Network 120 100 Number of Occurrences 80 60 40 20 0 20 25 30 35 40 45 50 55 Number of States The states are well distributed with a C.o.V 0.101

Normalized Path Length 4.5 WSR GLS-GPSR 4 DSR 3.5 Normalized Path Length Packets take longer 3 paths with WSR GLS sends packets only to 2.5 destinations that are successfully located 2 1.5 1 400 500 600 700 800 900 1000 Number of Nodes

But, E2E Delay is Lower! 70 WSR GLS-GPSR 60 DSR 50 End to End Delay (s) 40 WSR uses random walks 30 for discovery & dissemination => end- 20 to-end delay is smaller 10 0 400 500 600 700 800 900 1000 Number of Nodes

Discussion/Future Work q Weak State Routing also relates to þ DTN routing þ Unstructured rare object recall in P2P networks þ Distributed Hashing q Future work: þ Such extensions (especially DTNs) þ Theoretical analysis

Conclusion q Weak state is soft, updated locally, probabilistic and captures uncertainty q Random directional walks both for location advertisement and data forwarding. q WSR provides scalable routing in large, dynamic MANETs q WSR achieves high delivery ratio with scalable overhead at the cost of increased path length

Thank you Questions?