Skip to Main Content
With the increasing use of network of workstations (NOWs) as an alternative to huge parallel computers it has become essential to design high-performance interconnection networks for the communication between the nodes of these clusters. A large number of studies have been carried out to achieve this objective. Most of them propose a new technique that affects one of the parameters that characterize the interconnection network. These techniques are completely new or inspired in the techniques previously used in multiprocessor systems. The impact of the proposal is studied (in most cases using simulation), and an analysis is made of the effect of the new technique over the system performance versus those currently in existence. In this kind of study most of the network parameters are fixed and usually only a few parameters are varied. This paper presents a more general study of the interconnection network performance. This study consists in showing the effect of different design parameters over the network performance, and the interaction between them. This study would not be viable with the traditional techniques due to the number of simulations required. The alternative of the experimental design is used to carry out the study.