My current research interest, the High Performance Virtual Machines (HPVM) project, involved constructing advanced computing platforms from clusters based on commodity hardware (x86-based systems) and software (Windows NT).

Unlike many related projects, HPVM does not rely exclusively on commodity components. Rather, it exploits custom additions to existing systems. To date, these additions have been exclusively software. Our Illinois Fast Messages (FM) messaging layer and associated device drivers achieve communication performance that is several orders of magnitude faster than that achievable through more standard messaging layers such as those embodied in Winsock 2. On a cluster of PPro+WinNT systems using FM and our other HPVM technologies, we have demonstrated performance on parallel scientific applications superior to that achievable on mid-range parallel supercomputers. We are in the process of extending this same supercomputer-level performance to standard intranets.

My research goal is to advance HPVM technology with carefully-codesigned software and hardware working in unison to deliver performance far beyond that achievable by fast hardware or fast software alone. Furthermore, I plan to produce this new technology in the context of Windows NT and x86 computers to make HPVM as widely-available and universally-acceptable as possible.

August, 1997