next up previous index
Next: T1.3 Status and Future Up: Session T1. Computational Infrastructure Previous: T1.1 The Future of   Author Index


T1.2 Experience with a Large-scale Windows NT Cluster

Andrew A. Chien (UCSD & NCSA)

The rapid maturation of desktop and high volume technologies in microprocessors, high speed networks, and operating systems make commodity building blocks powerful elements for building supercomputing systems. Exploiting High Performance Virtual Machines and Fast Messages software technologies, we have built a series of Windows NT clusters:

All of these clusters utilize Fast Messages and High Performance Virtual Machines software combined with gigabit networks from Myricom to deliver supercomputer class communication and system performance. The HPVM III cluster was build in conjunction with the National Center for Supercomputing Applications and demonstrated in April 1998. It has aggregate capacities of 50 Gigaflops, 48 Gigabytes of DRAM, bandwidth, and 400 Gigabytes of disk storage. The core technology, experiences with building these clusters, performance on microbenchmarks and applications, as well as challenges and technical directions will be discussed.


next up previous index
Next: T1.3 Status and Future Up: Session T1. Computational Infrastructure Previous: T1.1 The Future of   Author Index
adass@ncsa.uiuc.edu