Study of a cluster-based parallel system through analytical modeling and simulation

Abstract

In this paper we present a new analytical model for cluster-based parallel systems based on Multi-Chain Open Queuing Network. The proposed model is general enough in terms of node design and clustering which can be extended to any parallel system. To have a good estimation of the model parameters, a benchmark suite has been applied to a real system and parameters calibration is performed based on actual measurements. In this study the hypercube topology was chosen for the interconnection network of the system. We present an analytical modeling and validate it by simulation. The QNAT is used for analytical modeling and the OMNet++ simulator has been used to carry out the simulations. Comparison of simulation and analytical results verifies validity of the model. System workload in simulations is generated using both analytical and experimental distributions and a close match with analytical results in both cases is observed. © Springer-Verlag Berlin Heidelberg 2005.