In recent years, more emphasis has been placed on the design and performance evaluation of processor and memory architectures than on I/O subsystems in parallel processors. As a result, a large gap exists between I/O access and processor and memory access times in these machines, causing I/O to become a prohibitive bottleneck. Presented are the results of a study conducted to evaluate the performance of parallel I/O on a massively parallel processor (MPP), consisting of a number of compute nodes and I/O nodes. A general MPP architecture is assumed. The number of nodes evaluated ranges from 16 to 128 with a variation in the ratio of compute nodes to I/O nodes. We calculate the network traversal and total processing times for I/O reads and writes while varying the I/O request rate, the request rate for non-I/O message traffic, and the request size. We also study the impact of non-I/O messages on the performance of I/O traffic and, conversely, how the I/O traffic affects the processor-processor messages. For I/O-only traffic, the results show performance advantages for I/O processing times as the number of I/O nodes in the system is increased, and as the I/O request rate increases. The system is scalable for the I/O loads considered in this study. Adding non-I/O traffic results in performance degradations ranging from 2% to up to an order or magnitude as the non-I/O message traffic increases to high loads of up to 4464 requests/sec. The scalability of this system may be limited by the I/O node saturation or considerable network contention. (ScalParSys)
By: Sandra Johnson Baylor, Caroline Benveniste, and Yarsun Hsu
Published in: RC20049 in 1995
LIMITED DISTRIBUTION NOTICE:
This Research Report is available. This report has been submitted for publication outside of IBM and will probably be copyrighted if accepted for publication. It has been issued as a Research Report for early dissemination of its contents. In view of the transfer of copyright to the outside publisher, its distribution outside of IBM prior to publication should be limited to peer communications and specific requests. After outside publication, requests should be filled only by reprints or legally obtained copies of the article (e.g., payment of royalties). I have read and understand this notice and am a member of the scientific community outside or inside of IBM seeking a single copy only.
Questions about this service can be mailed to reports@us.ibm.com .