Monday, April 02, 2001
11:00 AM
CSB 209
Mark Heinrich
Cornell U.
Hardware DSM Performance at Software DSM Cost
Emerging trends in commodity network technology coupled with key insights from academic research in active memory systems are leading toward the realization of hardware DSM on commodity clusters. We call the result of this convergence active memory clusters. After discussing the current state of the art in hardware DSM, clusters, and software DSM architectures, I will highlight the key differences between hardware and software DSM systems and show how these differences are rapidly disappearing in commodity systems---with the notable exception of the specialized memory controller present in hardware DSM systems. I will then discuss recent research results in active memory systems showing that uniprocessor performance can be improved with the inclusion of an active memory controller, indicating that it may become part of forthcoming commodity workstations. We contend that active memory support can be treated as an extension of the cache coherence protocol, and that an active memory controller contains the necessary functionality for building a hardware DSM machine. Coupled with enhancements in network technology and a small amount of software support, active memory clusters can achieve hardware DSM performance at software DSM cost.