Eric Anderson, University of California, Berkeley

Abstract:

We are looking at building large (100-1000 processor) clusters of machines using NT as the base operating system as a follow-on to the research we have done using unix in networks of workstations. We have determined that with appropriate access to sources, etc, we should be able to port all that we have done so far onto NT, but we are trying to figure out what additional things we will be able to leverage if we start thinking more from scratch than thinking as a port. The biggest advantages that have been identified for NT are the DCOM (distributed objects) layer, and the Wolfpack fault tolerance work.

Personally, I am looking at administration of NT to make sure that the work I have implemented on unix works on NT as well. While I believe that it will, I have not yet ported any of the software to NT, so do not have concrete proof that the architecture is sufficiently general to encompass NT.