geometric spanners for routing in mobile networks
DESCRIPTION
Geometric Spanners for Routing in Mobile Networks. Jie Gao, Leonidas Guibas, John Hershberger, Li Zhang, An Zhu. Motivation. Motivation: Efficient routing is difficult in ad hoc mobile networks. - PowerPoint PPT PresentationTRANSCRIPT
Geometric Spanners for Routing in Mobile
NetworksJie Gao, Leonidas Guibas, John Hershberger, Li Zhang, An
Zhu
MotivationMotivation:
Efficient routing is difficult in ad hoc mobile networks. Geographic forwarding, e.g., the Greedy Perimeter
Stateless Routing (GPSR) protocol, can be used with a location service.
GPSR is based on the Relative Neighborhood Graph (RNG) or the Gabriel Graph (GG) for connectivity.
Our approach: Restricted Delaunay Graph (RDG).
Combined with a mobile clustering algorithm. Good spanner in both Euclidean & Topological
distance. Efficient maintenance in a distributed setting.
Prior Work
Many routing protocols: Table-driven Source-initiated on-demand Greedy Perimeter Stateless Routing
(GPSR) by Karp and Kung, Bose and Morin.
Clustering in routing: Lowest-ID Cluster Algorithm by
Ephremides et al.
From Computational Geometry..
Graph Spanner G’ G Shortest path in G’
const optimal path in G Stretch factor
Delaunay Triangulation. Voronoi diagram. Empty-circle rule. Good spanner.
Voronoi cell
NodeDelaunay
Edge
Empty circle certifies the Delaunay edge
Construction of the Routing Graph
Assume the visible range = disk with radius 1.
1. Clusterheads. 2. Gateways.3. Restricted Delaunay
Graph on clusterheads and gateways.
Routing graph =RDG + edges from
clients to clusterheads.
u
v
A Routing Graph Sample
Select clusterheads
Clusterheads select gateways
RDG on clusterheads & gateways
Mobile Clustering Algorithm
1-level clustering algorithm: Lowest-ID Cluster algorithm.Hierarchical algorithm: proceed the 1-level clustering algorithm in a number of rounds. Constant Density Property for hierarchical clustering: # of clusterheads and gateways in any
unit disk is a constant in expectation.
Clustering Demo
Clusterheads
Clients
Disappearing ClusterheadsNew Appearing Clusterheads
Restricted Delaunay Graph
A RDG Is planar (no crossing edges). Contains all short Delaunay
edges (<=1).
RDG is a spanner Euclidean Stretch factor:
5.08. Topological spanner.
Routing graph is a spanner, too. Both Euclidean & topological
distance.
Short D-edgeLong D-edge
Maintaining RDG1. Compute local Delaunay triangulation.2. Information propagation.3. Inconsistency resolution.
a’s local Delaunay b’s local Delaunay
Maintaining GatewaysClusterheads maintain a maximal matching. Update cost = constant time per node.
(1) The original maximal matching between clients of two clusterheads.
(2) A pair of nodes become invisible.
(3) A node leaves the cluster.
(4) A new node joins the cluster.
Edges in matching
Edges in bipartite graph, not in matching
Quality Analysis of Routing Graphs
Optimal path length = k, greedy forwarding path length = O(k2), perimeter routing in the correct side = O(k2).
Greedy forwarding Perimeter routing
Simulation (Uniform Distribution)
•300 random points.
•Inside a square of size 24.
•Visible range: radius-2 disk.
•1-level clustering algorithm.
•RNG v.s. RDG under GPSR protocol.
•Static case only.
RNG
RDG
Simulation (Uniform Distribution)
Average path length
Maximal path length
Simulation (Non-uniform Distribution)
DiscussionScaling vs. spanner property Cannot be achieved at the same time.
Efficiency of clustering No routing table. Update cost: constant per node. Changes happen only when topology
changes.
Forwarding cost RDG: constant. RNG (or GG): Ω(n).
Conclusion
Restricted Delaunay Graph Good spanner. Efficiently maintainable. Performs well experimentally.
Quality analysis of routing paths Under greedy forwarding Under one-sided perimeter routing
Demo
client Edges in RDGEdges to connect clients to clusterheads
clusterheadgateway