Date: Tue, 05 Nov 1996 00:26:42 GMT Server: NCSA/1.5 Content-type: text/html Last-modified: Wed, 31 May 1995 20:57:45 GMT Content-length: 4310 Blizzard & Paradyn WWW Status Report

Blizzard and Paradyn: Infrastructure and Scalable Tools for Multi-Paradigm Parallel Computers

WWW Status Report as of January 1995


Wisconsin COW Wisconsin COW

Recent Accomplishments:

Last Update: January 1994; Contact Barton Miller (bart@cs.wisc.edu)

Background:

The national defense and other key sectors--such as education, health care, and commerce--increasingly benefit from and relying upon massive computational power. Parallel computers constructed from an array of conventional computers offer a cost-effective approach to buying CPU cycles. Although the cost advantage is clear, getting the computers to work together effectively requires better methods for:
  1. cooperatively sharing information, and
  2. identifying and mitigating "information traffic jams."

Goals:

This project addresses both concerns. Effective information sharing must permit programmers to tailor a program's sharing to the problem being solved--rather than being constrain by the limit options built into a system. Our system supports message passing (like postal mail), fine-grain shared memory (like reading over each other's shoulders), and hybrid combinations of the two. Our system can be implemented on low- to high-end parallel computers, so programs can be reused across a wide range of system. For more information on this part of the project, click here.

The second goal of project is to improve techniques for mitigating information traffic jams--what computer specialists call "performance debugging." Traditional tools for performance debugging have not scaled to large parallel machines because it is difficult to decide in advance what to measure and in what detail. Our tools transcend this problem by supporting "dynamic instrumentation" that allows measurement activity to adapt to the tool's current hypothesis for the information traffic jam's location. For more information on this part of the project, click here.

Participants & Support:

The principal investigators are Barton P. Miller, Mark D. Hill, James R. Larus, and David A. Wood in Computer Sciences at the University of Wisconsin.

This work is supported in part by Wright Laboratory Avionics Directorate, Air Force Material Command, USAF, under grant #F33615-94-1-1525 and ARPA order no. B550, NSF PYI/NYI Awards CCR-9157366, MIPS- 8957278, and CCR-9357779, NSF Grants CCR-9101035 and MIP- 9225097, DOE Grant DE-FG02-93ER25176, University of Wisconsin Graduate School Grant, Wisconsin Alumni Research Foundation Fellowship and donations from A.T.&T. Bell Laboratories, Digital Equipment Corporation, Sun Microsystems, Thinking Machines Corporation, and Xerox Corporation. Our Thinking Machines CM-5 and Cluster of Sun Workstations were purchased through NSF Institutional Infrastructure Grant No. CDA-9024618 with matching funding from the University of Wisconsin Graduate School.

Last Update: January 1994; Contact Barton Miller (bart@cs.wisc.edu)