How Moderate-Sized RIS C-Based SMPs Can Outperform Much Larger Distributed Memory MPPs

摘要

Historically, comparisons between computer systems were based primarily on theoretical peak performance. Today, comparisons based on delivered levels of performance are frequently used. This, of course, raises a whole host of questions concerning methodology. From the standpoint of the user, delivered performance frequently refers to how fast a job runs. However, is it reasonable to base this measurement on running the same algorithm on all of the computers. When comparing some combination of mainframes and vector supercomputers, the answer is probably yes. The same holds true when comparing the performance of large distributed memory MIMD MPPs. However, when comparing the algorithms of choice used on these two classes of platforms, one frequently finds that the algorithms are quite different. Furthermore, the amount of work (the number of FLOPS) associated with each algorithm can also be quite different. This implies that for an MPP to have the same level of delivered performance as the mainframes and the vector supercomputers, it must have a significantly greater level of performance when measured in terms of FLOPS. Recent advances involving moderate sized RISC based SMPs have allowed us to solve this problem. The net result is that for some problems a 128 processor Origin 2000 can deliver levels of performance that might require the use of a 500 processor MPP using more traditional approaches.