A Practical Approach to QoS Routing for Wireless Networks Teresa Tung, Zhanfeng Jia, Jean Walrand WiOpt 2005—Riva Del Garda
Outline Problem: clustering Assumptions: routing algorithm Analysis: simple models Analysis: simulations
Scenario Routing over ad-hoc wireless networks Goal: Discover the diverse paths Small area, use shortest path Uniform demand, shortest path admits most flows Demand between few s-d pairs, use diverse paths to increase capacity
Observation on Interference Interference –Area effect –Not a link effect Routing choices –Over areas –Not over links TxIntfx
Related Work Theoretical Approach Gupta Kumar Thiran Practical Fixed transmission radius Routing algorithms
Clustering: Motivation Clustering makes sense for dense networks Each node sees roughly the same info
Clustering: Motivation Clustering makes sense for dense networks Each node sees roughly the same info
Clustering: Motivation Clustering makes sense for dense networks Each node sees roughly the same info
Clustering: Motivation Clustering makes sense for dense networks Each node sees roughly the same info
Costs Cost of flat routing –No point in all nodes reporting –Reduction in control messages –Limited loss of information Cost of clustering –Restrict possible paths –Use more network resources
Outline Problem: clustering Assumptions: routing algorithm Analysis: simple models Analysis: simulations
Routing granularity Comparison of routing strategies over a flat network shows little improvement Scheme –Shortest path within clusters –OSPF at the cluster level –Measurement –Admission Control
Routing Source Dest
Routing
Routing: Measurement Measure the available resources in a cluster Use a representative node per cluster Given the link speed Measure the fraction of time that the channel is busy –Transmitting/Receiving –Channel busy The fraction of idle time x link speed gives an upper bound on residual capacity
Routing: OSPF weights Estimate residual capacity Shortest feasible path Most probable path Residual capacity
Routing: Admission Control For inelastic flows require a rate F Trial flow of same rate F for period t Trial packets served with lower priority Admit if all trial packets received Otherwise busy e Admitted Trial high
Routing Assumptions Shortest path within clusters Resource estimates via measurements OSPF based scheme at the cluster level Admission control
Outline Problem: clustering Assumptions: routing algorithm Analysis: simple models Analysis: simulations
Clustering: Analysis Model Continuous plane (dense network) Compare routes over an idle network Grid clustered Compare –Length –Self interference –Diversity
Compare # hops Clustering: Length
Path length: grid size
Path length: grid = 2r
Clustering: Self-Interference Unit disk model, interference radius Self-interference for shortest path
Clustering: Self-Interference Midpoint on II –From II –From I and III each Decreasing in grid size
Clustering: path diversity
Cost of Flat Routing N nodes over area A=ar x ar where r tx radius C=(a/g)^2 clusters of size gr x gr Average hops between nodes L Average hops across cluster < gsqrt2 Flat routing LN 2 Clustered routing (gc1+c2L)C 2
Outline Problem: clustering Assumptions: routing algorithm Analysis: simple models Analysis: simulations
Outline Problem Argument for clustering Routing scheme Simulation results
Simulations Matlab Algorithms Global OSPF Event driven OSPF Event+clustered OSPF 100 nodes, vary density Mesh topology (5x5) Random topology (3x3,4x4)
Clustering: Shortest Path
Simulations: Admission Ratio Mesh over a 5x5 Grid Random over a 3x3 Grid
Simulations: Max capacity s-d Mesh over a 5x5 Grid Random over a 3x3 Grid
Simulations: Average path length Mesh over a 5x5 Grid Random over a 3x3 Grid
Simulations: Path length for fixed s-d pair
Simulations: Path Diversity
Simulations: ave # routes s-d Mesh over a 5x5 Grid Random over a 3x3 Grid
Conclusion Cost of clustering: 20% loss in admit ratio Path length Self-interference Path diversity www-inst.eecs.berkeley.edu/~teresat