1 Oblivious Routing in Wireless networks Costas Busch Rensselaer Polytechnic Institute Joint work with: Malik Magdon-Ismail and Jing Xi
2 Outline of Presentation Introduction Network Model Oblivious Algorithm Discussion Analysis
3 Routing: choose paths from sources to destinations
4 Edge congestion maximum number of paths that use any edge Node congestion maximum number of paths that use any node
5 Length of chosen path Length of shortest path Stretch= shortest path chosen path
6 Oblivious Routing Each packet path choice is independent of other packet path choices
7 Path choices: Probability of choosing a path:
8 Benefits of oblivious routing: Appropriate for dynamic packet arrivals Distributed Needs no global coordination
9 Related Work Valiant [SICOMP’82]: First oblivious routing algorithms for permutations on butterfly and hypercube butterflybutterfly (reversed)
10 d-dimensional Grid: Lower bound for oblivious routing: Maggs, Meyer auf der Heide, Voecking, Westermann [FOCS’97]:
11 Azar et al. [STOC03] Harrelson et al. [SPAA03] Bienkowski et al. [SPAA03] Arbitrary Graphs: constructive Racke [FOCS’02]: existential result
12 Hierarchical clustering Approach:
13
14 At the lowest level every node is a cluster
15 source destination
16 Pick random node
17 Pick random node
18 Pick random node
19 Pick random node
20 Pick random node
21 Pick random node
22 Pick random node
23
24 Adjacent nodes may follow long paths Big stretch Problem:
25 An Impossibility Result Stretch and congestion cannot be minimized simultaneously in arbitrary graphs
26 Each path has length paths Length 1 Source of packets Destination of all packets Example graph: nodes
27 packets in one path Stretch = Edge congestion =
28 1 packet per path Stretch = Edge congestion =
29 Contribution Oblivious algorithm for special graphs embedded in the 2-dimensional plane Constant stretch Small congestion degree Busch, Magdon-Ismail, Xi [SPAA 2005]:
30 Embeddings in wide, closed-curved areas
31 Our algorithm is appropriate for various wireless network topologies Transmission radius
32 Basic Idea source destination
33 Pick a random intermediate node
34 Construct path through intermediate node
35 Stretch = Previous results for Grids: Busch, Magdon-Ismail, Xi [IPDPS’05] For d=2, a similar result given by C. Scheideler
36 Outline of Presentation Introduction Network Model Oblivious Algorithm Discussion Analysis
37 Network Surrounding area
38 space point space point Perpendicular bisector geodesic
39 space point space point
40 Area wideness:
41 space point graph node Coverage Radius : maximum distance from a space point to the closest node
42 there exist For all pair of nodes Shortest path length: Euclidian distance:
43 Consequences of (max transmission radius in wireless networks) edge Max Euclidian distance between adjacent nodes
44 Consequences of nodes Min Euclidian Distance between any pair of nodes:
45 Small and large Good Network embeddings: Suppose they are constants
46 Outline of Presentation Introduction Network Model Oblivious Algorithm Discussion Analysis
47 Every pair of nodes is assigned a default path default path Examples: Shortest paths Geographic routing paths (GPSR)
48 The algorithm source destination
49 geodesic Perpendicular bisector
50 Pick random space point
51 Find closest node to point
52 default path default path Connect intermediate node to source and destination
53 Outline of Presentation Introduction Network Model Oblivious Algorithm Discussion Analysis
54 Consider an arbitrary set of packets: Suppose the oblivious algorithm gives paths:
55 We will show: optimal congestion
56 Theorem: Proof: Consider an arbitrary path and show that:
57 default path default path shortest path
58 we show this is constant when default paths are shortest paths
59 Default path (shortest) Similarly:
60 shortest path
61 For constants: End of Proof
62 Theorem: Proof: Consider some arbitrary node and estimate congestion on Expected case: denotes
63 Deviation of default paths: maximum distance from geodesic geodesic
64 Consider some path from to
65 the use of depends on the choice of space point one choice
66 another choice
67 cone affecting If you choose node in the cone the respective path may use
68 cone affecting If you choose node outside the cone the respective path does not use
69 cone affecting Segment of space points affecting
70 Probability of using node : cone affecting
71 It can be shown that: constant
72 for simplicity assume:
73 : constants
74 Divide area into concentric circles
75 Max Euclidian distance between any two nodes = Longest path has at most nodes
76 Maximum ring radius
77 = number of packets that can affect = number of paths that use Ring We will bound
78
79 Expected congestion:
80 We have proven we prove next
81 we showed earlier
82 Similarly, each packet that affects traverses distance at least
83 Area
84 Total number of nodes used Area
85 Average node utilization Area
86 #nodes in area = Area
87 Average node utilization average node utilization
88 We have proven:
89 Considering all the rings: End of Proof
90 Recap Constant stretch Small congestion We presented a simple oblivious algorithm which has: when the parameters of the Euclidian embedding are constants
91 Outline of Presentation Introduction Network Model Oblivious Algorithm Discussion Analysis
92 Holes
93 Arbitrary closed shapes there is no