The goal of this project is two-fold. First, it seeks to study the communication characteristics of applications that can benefit from the computing power, very large memory, and high-speed communication of parallel computing systems. Within this context measurements are being takel of the traffic generated by both regular and irregular scientific applications, applications written in a variety of programming paradigms, data-mining applications and applications with real-time traffic demands. Second the project seeks to use the information gathered from applications to guide the designs for interconnect devices and programmable communication coprocessors that efficiently support multiple classes of applications and multiple programming paradigms. The education plan addresses two important issues in experimental computer science: First it addresses the issue to teaching a subject area, such as high performance computing, that is technology-driven and where the relevance of knowledge and information can have a very short life span. Second, it addresses the issue of teaching the methodology of experimentation and in particular the process of designing experiments.
