Guaranteed Greedy Routing in Overlay Networks
1 like759 views
The document discusses guaranteed greedy routing in overlay networks. It proposes constructing an overlay network using a Nearest Neighbors Convex Set (NNCS) that guarantees nodes will always have a neighbor closer to the destination. This allows for greedy routing to make progress at each hop. The NNCS is constructed by dividing the virtual space around each node into half-spaces defined by its neighbors. Various applications like multicast trees, service discovery, and spatially-bound querying can take advantage of the NNCS approach. Evaluation shows the NNCS outperforms Pastry in terms of relative path stretch.
Guaranteed Greedy Routing in Overlay Networks
- 1. Conference on Future Internet Communications 2013 Guaranteed Greedy Routing in Overlay Networks Dragan Milic and Torsten Braun Universität Bern, Switzerland braun@iam.unibe.ch, cds.unibe.ch
- 2. Overview > Introduction — Motivation — Greedy Routing in Virtual Spaces — Greedy Routing Failure > Overlay Network Construction for Greedy Routing — Nearest Neighbours Convex Set (NNCS) — Algorithm for NNCS Maintenance — NNCS Robustness — Routing Optimization > NNCS Applications — Building of Multicast Trees — Service Discovery — Spatially Bound Query Flooding — Network Optimization / QoS > Evaluation: Relative Path Stretch > Conclusions CFIC, Coimbra, May 16, 2013 Torsten Braun: Guaranteed Greedy Routing in Overlay Networks 2
- 3. Motivation > Overlay network construction does often not consider underlying network topology. > Overlay network routing protocols — optimize number of hops along the path, but — disregard paths’ RTT stretch = actual RTT on the path through the overlay network RTT along the optimal path in the underlying network > Proposal: apply greedy routing in virtual spaces, where distance between nodes is proportional to their RTT CFIC, Coimbra, May 16, 2013 Torsten Braun: Guaranteed Greedy Routing in Overlay Networks 3
- 4. CFIC, Coimbra, May 16, 2013 Torsten Braun: Guaranteed Greedy Routing in Overlay Networks 4 Positioning Hosts in a Virtual Space Host 1 Host 2 Host 8 Host 3 Host 5 Host 7 Host 4 Host 6 Physical Network Virtual Space Host 1 Host 8 Host 2 Host 3 Host 7 Host 4 Host 5 Host 6 x y
- 5. Greedy Routing in Virtual Spaces > Next hop = neighbour with minimal distance to destination CFIC, Coimbra, May 16, 2013 Torsten Braun: Guaranteed Greedy Routing in Overlay Networks 5
- 6. Greedy Routing Failure > Problem — There might not always be a neighbour closer to the destination. > Solutions — For geographical routing in mobile ad-hoc networks: backup routing modes, e.g., using right-hand rule in GPSR — For overlay networks: construct an overlay network that ensures to have always a neighbour closer to the destination CFIC, Coimbra, May 16, 2013 Torsten Braun: Guaranteed Greedy Routing in Overlay Networks 6
- 7. Overlay Network Construction for Greedy Routing CFIC, Coimbra, May 16, 2013 Torsten Braun: Guaranteed Greedy Routing in Overlay Networks 7 > Delaunay triangulation — can ensure existence of closer neighbours towards destination, — but distributed approach is complex. > Solution — Nearest Neighbors Convex Set (NNCS) is based on dividing the virtual space into half spaces. Current Node Neighbour Node Half Space
- 8. Nearest Neighbours Convex Set (NNCS) > Construct half spaces for all neighbours of a node > NNCS — is a convex set defined by intersection of all half spaces. — guarantees that greedy routing makes progress at each node. CFIC, Coimbra, May 16, 2013 Torsten Braun: Guaranteed Greedy Routing in Overlay Networks 8 Current Node Neighbour Node Other Nodes Half Space
- 9. Algorithm for NNCS Maintenance CFIC, Coimbra, May 16, 2013 Torsten Braun: Guaranteed Greedy Routing in Overlay Networks 9
- 10. NNCS Robustness > Problem — In case of churns or failures, NNCS may lead to routing failures or isolated nodes. > Solution — Add second NNCS layer by constructing a NNCS without considering the nodes in the first layer of the NNCS CFIC, Coimbra, May 16, 2013 Torsten Braun: Guaranteed Greedy Routing in Overlay Networks 10
- 11. Routing Optimization CFIC, Coimbra, May 16, 2013 Torsten Braun: Guaranteed Greedy Routing in Overlay Networks 11 > Problem — NNCS selects close neighbours, which leads to many hops. — Fisheye overlay network with a mix of close and distant neighbours, but greedy routing using a Fisheye overlay network does not guarantee progress. > Solution — maintain NNCS and Fisheye overlay network sets — select next hop making most progress from both node sets D. Milic, T. Braun: “Fisheye: Topology aware choice of peers for overlay networks”, 34th IEEE Local Computer Networks, 2009
- 12. Applications > Building of Multicast Trees > Service Discovery > Spatially Bound Query Flooding > Network Optimization / QoS CFIC, Coimbra, May 16, 2013 Torsten Braun: Guaranteed Greedy Routing in Overlay Networks 12
- 13. Building of Multicast Trees > Reverse path multicast routing > Node R forwards a received multicast message to all nodes that would use R for routing a unicast message to the source CFIC, Coimbra, May 16, 2013 Torsten Braun: Guaranteed Greedy Routing in Overlay Networks 13
- 14. Service Discovery CFIC, Coimbra, May 16, 2013 Torsten Braun: Guaranteed Greedy Routing in Overlay Networks 14 > Problem — Service discovery is usually based on flooding. — Efficient flooding mechanisms > Solution — Forwarding of query message along reverse path Query Source End Systems with Query Source in their NNCS Other Nodes Flooding Path
- 15. Spatially Bound Query Flooding > Queries can easily bound within a certain area, e.g., a hyper ball, to ensure RTT between query source and server. CFIC, Coimbra, May 16, 2013 Torsten Braun: Guaranteed Greedy Routing in Overlay Networks 15 Query Source End Systems with Query Source in their NNCS Other Nodes Flooding Path
- 16. Network Optimization / QoS > Paths fulfilling RTT / QoS requirements must be within an ellipsoid around source and destination. > Ellipsoid as bounding area for path to be selected. CFIC, Coimbra, May 16, 2013 Torsten Braun: Guaranteed Greedy Routing in Overlay Networks 16
- 17. Relative Path Stretch > NNCS evaluation using OMNET++ > Different data sets — PlanetLab data for 217 overlay nodes — KING data for 462 overlay nodes > Comparison of NNCS using different virtual space dimensions with Pastry CFIC, Coimbra, May 16, 2013 Torsten Braun: Guaranteed Greedy Routing in Overlay Networks 17
- 18. Conclusions > NNCS enables greedy routing in overlay networks using virtual space embedding. > Various applications can take advantage of NNCS. > NNCS based approach outperforms Pastry. CFIC, Coimbra, May 16, 2013 Torsten Braun: Guaranteed Greedy Routing in Overlay Networks 18
- 19. Thanks for your Attention ! > cds.unibe.ch > slideshare.net/torstenbraun > braun@iam.unibe.ch Torsten Braun: Guaranteed Greedy Routing in Overlay Networks CFIC, Coimbra, May 16, 2013 19