Oblivious Routing in Wireless Mesh Networks

1
Oblivious Routing in Wireless Mesh Networks
http//pdos.csail.mit.edu/roofnet/doku.php?idmap

• ICC 08
• May 22
• Jonathan Wellons and Yuan Xue
• Vanderbilt University

2
Outline

• Wireless Mesh Networks
• Congestion Model
• Oblivious Routing and Formulation
• Simulation
• Results
• Future Research

3
Wireless Mesh Networks
4
Interference Model
Interference Set of (i, j) all edges that cannot
intelligibly transmit at the same time as (i, j)
? (i, j).
5
Definitions Routing

• Network G (V, E)Demands D dijFlow
  source/destination/quantityPathway Sequence of
  nodesRouting a set of pathways

6
Optimal Routing

• Minimize max congestion
• Suppose dEF dJK 1

7
Congestion Metric

• Congestion on busiest interference set
• Da,b demand from a to b
• fa,b(i,j) portion of flow from a to b that
  crosses link (i,j)
• Capi,j capacity of link ij
• Then the maximum congestion in the network is

8
Minimizing Congestion(Demand-Aware/Oracle)

• Find a routing, f, that has the least congestion.
• An easy LP problem
• Minimize r, such that
• F is a routing, demands are met and congestion at
  all edges is r.

9
Oblivious Routing

• Demand is unknown
• Routing are evaluated based on worst-case demand.
• Oblivious routing will have the best worst
  case.
• Demand can be very erratic See CRAWDAD trace.

10
Quality of a Routing

• How do we say one routing is a better oblivious
  routing than another?
• Flow(e, f, D) is the flow on e, in routing f with
  demands D.

11
Quality of a Routing

• Let D be the set of all possible demands. Let
  OPT(D) be the congestion achieved by the optimal
  routing. Then let

• We Seek

12
Oblivious Formulation

• Problem is non-linear A maximization embedded in
  a minimization problem.
• Use Duality to convert into a single LP problem
• Produces a single, polynomial-sized set of
  equations
• Solve with off-the-shelf (or off-sourceforge) LP
  solver
• Details in paper (Oblivious Routing for Wireless
  Mesh Networks)

13
Simulated Demand
14
Simulated Topology
15
Results
16
Future Work

• Multiple Channels
• Partially-known Demand
• More simulations setups and comparisons with more
  algorithms
• Understanding the impact of topology and demand
  erraticity on oblivious performance.

17
Thanks

• IEEE ICC 2008
• Vanderbilt University