Performance analysis of a non-homogeneous GPS system under Markov modulated fluid traffic

This paper presents a near-exact analysis of the generalized processor sharing (GPS) scheduler under Markov modulated fluid sources (MMFS). Due to the complexity of the GPS system, all previous works are based on so called bounding approach providing upper bound using deterministic analysis or decomposition techniques. These bounds are in general quite loose and result in overly conservative resource provisioning. In this paper, a novel approach is used to obtain near-exact solutions for the queue length distribution of each session. The analysis applies to the general case of N parallel queues with multi-class non-homogeneous sources and non-rate-proportional weight setting of sessions. Numerical results for traffic engineering under various system and traffic parameters are reported.