Modeling and analysis of multiprocessor architectures (W.M. Zuberek)

Multiprocessor architectures are composed of a number of processors, each with its (local) memory, and an interconnection network. Because of very simple representation of concurrency and synchronization, timed Petri net models seem to be well suited to modeling and evaluation of such architectures. However, accurate net models of multiprocessor architectures become quite complicated, and then their analysis is a nontrivial task. High-level Petri net models, and in particular colored Petri nets, can be used to simplify the model by "folding" all similar components of the system into a single one. The simplification of the model, however, typically increases the complexity of its analysis.

The objective of this research is to develop timed Petri net models of a wide range of multiprocessor architectures, and in particular, multithreaded multiprocessors. Although many examples refer to a 16-processor system connected by a two-dimensional torus-like network, as shown in Fig.1, they might as well refer to a 16-processor system with a hypercube-type connecting network, as shown in Fig.2, or other types of interconnections. The concept of performance equivalence explains the performance similarities between such two systems, as well as between many others, for example, a 64-processor system connected by a three-dimensional torus-like network and a 64-processor system connected by a hypercube network.

Fig.1. 16-processor system connected by a 2-dimensional torus-like network.

Fig.2. 16-processor system connected by a hypercube-type network.

Specific projects in this area include:

Up to Main Page Next Page

Copyright by W.M. Zuberek. All rights reserved.
Revised: 2004.07.25 :