This research addresses the problem of providing efficient, protected communication in an SMP cluster. Traditional approaches use either custom hardware of system-calls. This paper analyzes a new approach that uses an idle SMP processor to run a message proxy, a communication process that provides protected access to the network. We implement a message proxy and measure in detail the resulting communication overheads between a pair of IBM Model G30 SMPs. We derive a performance model that shows that cache-miss latency within an SMP influences message proxy performance significantly. We compare the performance of message proxies with custom hardware and system calls. Simulations of a suite of ten parallel applications demonstrate that message proxies match the performance of custom hardware for three of the ten applications, and are between 10-30% slower for the other seven applications. A direct cache-update mechanism to reduce cache misses improves the performance of message proxies on communication-intensive programs by 7-25%. Using system-calls results in the poorest performance. We conclude that the message proxies provide a viable alternative to custom hardware for protected communication.
By: Beng-Hong Lim, Philip Heidelberger, Pratap Pattnaik and Marc Snir
Published in: RC20522 in 1996
This Research Report is not available electronically. Please request a copy from the contact listed below. IBM employees should contact ITIRC for a copy.
Questions about this service can be mailed to reports@us.ibm.com .