Our first distinguished speaker for the 1996-1997 Series is scheduled
for Wednesday, November 20, 1996 in the McDonnell Douglas Auditorium
beginning w/refreshments at 10:30am in the lobby. The talk will begin
promptly at 11:00 and will last approximately an hour.
See http://www.ics.uci.edu/talks.html for a map to the building or
call (714)824-7403 if you have any questions.
Network of PCs as High-Performance Servers: The SHRIMP Approach
We are moving into a new era of inexpensive, high-performance
computing using networks of PCs, workstations and servers, and moving
away from computing infrastructures built around traditional
mainframes, custom-designed parallel computers, and specialized
servers. The reason for this new era is simple: Special-purpose
computers are very expensive, and unable to track the rapid
development of hardware and software technologies. The approach to
using entire systems to construct scalable systems, on the other hand,
can track closely the exponential increase in all aspects of the
technology base and reduce the cost of a multicomputer by an order of
Networks of commodity PCs have the potential to provide extremely
cost-effective computing and information services. However, they will
fulfill their promise only if their system performance is truly
competitive with commercial multicomputers or multiprocessors. To
achieve this goal, the SHRIMP project at Princeton attacks several
research challenges including low-latency and high-bandwidth network
communication, efficient message-passing mechanisms and shared memory
system, parallel storage system, and support for parallel graphics.
This talk first addresses several issues and challenges in building
high-performance servers using a network of commodity computer
systems, gives an overview of the SHRIMP project, and then describes
recent research results.