An Interview with Robert Pennington

Robert Pennington is the Windows NT Cluster Group's team leader at the National Center for Supercomputing Applications (NCSA)—the group that saw the growing importance of NT and Intel and built the 192-processor NT cluster. Pennington, who reminded me that the NT cluster is a research project, commented on the project.

What did you like about the project's implementation?

The implementation of this project leveraged a great deal of off-the-shelf hardware and software with a computer science research project to produce a very good system for computational research. The implementation of the NT cluster at NCSA clearly shows the ability of the National Computational Science Alliance to bring computer science research projects together with computational researchers who need these capabilities to work on their science. We have been able to scale up High Performance Virtual Machines (HPVM), a well-done and well-supported project from Andrew Chien's research group, with Intel-based computational platforms from HP and Compaq using Myricom's Myrinet interconnects. The resulting system, with the Intel-based architecture and the NT OS, has surpassed many of our expectations.

What did you dislike about the project's implementation?

We disliked really nothing about this project's deployment. The results have been even better than we anticipated.

What would you have done differently on the project?

We would've done nothing differently. The NT cluster is a research project. We are concerned only with whether this project is of benefit to our users. The answer to that concern has been an unequivocal Yes.

What advice can you give your peers?

We have worked with standard systems from the commercial hardware and software vendors as much as possible to be able to supply the NCSA users with state-of-the-art systems. Because the expected useful lifetime of the systems for us is only 12 to 18 months, this strategy is essential, as there isn't time between hardware generations to create—and recreate—entirely custom solutions on a large scale in such a short time. We have established and maintained excellent relationships with the vendors and researchers, both of which have been very important.