On the advantages of optimal end -to -end QoS budget partitioning

Hyunjoon Cho, Purdue University

Abstract

We investigate the problem of optimal end-to-end QoS budget partitioning to quantify the advantage of having a non-uniform allocation of end-to-end QoS requirement over the links in a path. We formulate an optimization problem that provides a unified framework to study static QoS budget allocation combined with network dimensioning and routing. Through the analysis of its optimality conditions, we examine the underlying mathematical structure for the model and give an economic interpretation on the optimal rules. Two aspects of the advantages of optimal partitioning are shown: reducing the construction cost in network dimensioning and increasing revenue in network operation. In the context of network dimensioning, we show that optimal partitioning can bring large cost reductions as compared with equal partitioning. Along with its advantage in network dimensioning, we also find that optimal partitioning brings a significant revenue increase in network operation with given capacity. More importantly, optimal partitioning improves robustness in the presence of failed components and fairness when the traffic demand is different from the forecast, which had not been observed in previous work. We show these advantages of optimal partitioning with some numerical results based on theoretical analysis. The concept of effective bandwidth is introduced to deal with both the QoS constraint and link blocking. We make an observation that the effective bandwidth model for the M/M/1 queueing system has an interesting property that the delay allocation can be done independently of traffic and routing. Based on the decomposition property of the M/M/1 case, a fast heuristic partitioning scheme is proposed for arbitrary traffic models, which can greatly reduce the computation effort. We then show that the heuristic gives a near-optimal partitioning for network dimensioning and that it can improve network robustness in network operation with given capacity. We also examine the effect of optimal QoS budget partitioning on routing. We analyze the relation between optimal partitioning and optimal routing rules and show that optimal partitioning improves the effect of rerouting on maximizing revenue under traffic change scenarios.

Degree

Ph.D.

Advisors

Shroff, Purdue University.

Subject Area

Electrical engineering

COinS