Massive Memory Means Massive Performance

Abstract:

While most of the computing world has been exploring the uses of parallelism to increase computational speed, we at Princeton have been examining methods of using large amounts of semiconductor memory to achieve the same end. We have demonstrated that time-space tradeoffs can be exploited across a wide range of applications to speed up computational tasks. On some tasks, our 128 megabyte VAX will outperform even the fastest supercomputers. These time-space tradeoffs coupled with rapidly falling semiconductor memory prices make Massive Memory systems economically inevitable. This paper documents performance results from our 128 megabyte testbed machine, and goes on to investigate promising application domains for Massive Memory computing systems.