Connection-wise end-to-end performance analysis of queueing networks with MMPP inputs

A systematic method for link/connection-wise end-to-end performance evaluation in queueing networks receiving heterogeneous Markov-modulated Poisson processes (H-MMPPs) is proposed. The method consists of (i) connection-wise nodal performance analysis; (ii) tagged departure process analysis; and (iii) moment matching. For the tagged departure process of an H-MMPPs/G/1 queue, we propose two decomposition schemes to approximate the output process of a tagged traffic stream which is mixed with other traffic streams. A moment matching method is further proposed to emulate the tagged output process as a two-state MMPP. The adjacent down-stream node along a reference connection can be then modeled as an H-MMPPs/G/1 queue. Recursively performing (i)–(iii), the end-to-end performance of a reference connection is obtained. The methodology developed in this paper can be applied to packet-switched high-speed networks, especially to asynchronous transfer mode (ATM) networks.