Performance prediction of wormhole switching in hypercubes with bursty traffic pattern

Wormhole switching has been extensively deployed in the current-generation of high-end parallel systems. The performance of wormhole switching in hypercubes has been primarily investigated under the assumption that traffic load follows a Poisson arrival process. However, many studies have shown that traffic in parallel computation environments can exhibit a high degree of burstiness and the Poisson arrival process is unable to model the behaviour of bursty traffic. This paper proposes a new analytical model for wormhole-switched hypercubes under bursty traffic. Simulation experiments demonstrate that the proposed model exhibits a good degree of accuracy under various operating conditions.