|Volume 7, Issue 1 -
CRPC Announces an Affiliated Site at the University of Illinois
The CRPC is proud to announce the establishment of an affiliated site at the University of Illinois at Urbana-Champaign. The site consists of the Pablo research group in the Department of Computer Science.
"We are delighted to join in the CRPC's mission of advancing the state of high-performance computing software" said Dan Reed, a professor of computer science who leads the group. "Not only has the CRPC created an extraordinary environment for distributed research collaboration, the community of CRPC researchers is at the cutting edge of high-performance computing research. Joining the CRPC will solidify and facilitate our ongoing collaborations with Rice, Caltech, and other CRPC sites."
The University of Illinois is the second affiliated site for the CRPC-- the first site at the University of Maryland was announced in the January 1994 issue of Parallel Computing Research. Affiliated sites work with one of the five major research thrusts and with associated groups including corporate affiliates. In addition, members of these sites attend the annual CRPC Research Symposium and technical steering committee meetings and participate in major conferences where the CRPC is represented.
The Pablo group's primary research focus is the capture and analysis of performance data from massively parallel computer systems, with the goal of understanding the interaction of architecture, system software, and applications. The primary vehicle for this work is the Pablo performance analysis environment, funded by ARPA, which provides portable performance data capture and presentation tools for scalable parallel systems. The Pablo performance analysis software has been distributed to academic sites throughout the United States and has been licensed to Intel, as the basis for the performance analysis tools on the Intel Paragon XP/S.
Specific collaborations with the CRPC include work with Ken Kennedy, John Mellor-Crummey and Vikram Adve at Rice University. The goal of this work is to make it possible for programmers to pose performance questions in terms of the machine-independent programming model used in the Fortran D language and to use these questions to acquire the performance data needed to produce performance analyses and visualizations that answer these questions in ways easily related to the user's program. Individually, the Pablo and Fortran D systems provide sophisticated code analysis facilities. However, each system addresses a different component of the overall problem of providing a productive software development environment for parallel computer systems. To provide performance feedback meaningful to a programmer writing in an abstract parallel language such as Fortran D, performance analysis tools must exploit information about the translation from the high-level source language to the low-level parallel code.
Using the Pablo software, the group is also working on high-performance input/output and integration of the Pablo environment with data parallel languages like Fortran D and HPF. As part of an NSF Grand Challenge group, the Pablo group is collaborating with the Caltech Concurrent SuperComputing (CSCC) facility (which is supported in part by the CRPC) to develop performance instrumentation extensions that allow the unobtrusive capture and analysis of input/output access patterns and latencies from a suite of parallel scientific applications. The goal is to understand the input/output demands of these codes and use that data to guide the design of flexible file system software.
CRPC Director Ken Kennedy stated, "The site at the University of Illinois is a welcome addition to our consortium. The collaboration between the CRPC and Dan Reed's group arose from a common objective to provide parallel users with an effective programming environment for High Performance Fortran and other emerging parallel programming languages. We believe that this collaboration will make it possible for the CRPC to better serve the entire high-performance computing community."
Table of Contents