Modelling And Analyzing Parallel Applications In Highly Distributed Systems

Nowadays, the use of cluster systems is one of the most cost-effective solutions for large-scale scientific applications. Designing and developing a system of this magnitude is a very expensive and complex task. This explains the need to estimate, in advance, the performance of new systems. Generally, it can be achieved by applying efficient techniques for evaluating and predicting the impact of distributed applications on the overall system performance. However, due to the complexity of the interactions among I/O devices, caches and network interfaces, the evaluation is often difficult to perform. The main aim of this work is to provide a method to model and analyze the impact of distributed applications on large clusters, providing enough flexibility to use different architectural configurations.