Berkeley Lab Buys 160-Processor Cluster Computer to Advance Scientific Computing & Research

In what could be a glimpse into the future of high-performance computing, Berkeley Lab Computing Sciences will buy and operate a 160-processor cluster computer to assess whether such system can meet the day-to-day production demands of a scientific computing center.

Clusters are assemblies of commodity computers designed and networked to operate as a single system. By using off-the-shelf components, clusters can provide a cost-effective balance between price and computer performance. To date, most large scientific commodity clusters are used more for specific research applications than general-purpose production resources for computational science.

“There’s been a lot of hype about clusters over the past few years in the world of high-performance computing, but clusters are just now coming into their own,” said NERSC Division Director Horst Simon. “Our goal with this new system is to see how well a commercially built cluster can perform in a demanding scientific environment. The cluster architecture could well be the supercomputer of the future and it behooves us to carefully and thoroughly test such a system.”

The primary goal in purchasing the cluster system is to provide a system testbed for NERSC and other staff to explore the applicability of such a system to support a highly parallel, numerically intensive workload. NERSC is interested in evaluating cluster architectures as an option for procurement of large production computing systems in the future.

The second objective is to provide a computational resource to strategic Berkeley Lab projects and campus collaborations which require significant computational resources while requiring only limited technical support.

The cluster, to be purchased from IBM, is expected to be delivered by March 2001. It will be named “alvarez” after LBNL Nobel Laureate Luis Alvarez.

The system is being purchased using University of California-Directed Research and Development (UCDRD) funds, which are provided to the Laboratory by the University of California for research activities or research-related activities at the Laboratory.

One of the stated reasons for moving NERSC to Berkeley Lab was to better integrate computational science into LBNL research programs. While the new cluster is expected to support a limited number of users, the experience in using a cluster to support Lab research could lead to more extensive, cost-effective computational support in the future. Lab management will develop a procedure for allocating time on the cluster.

Once the system passes its acceptance tests and is running, NERSC will have priority access to the cluster. About half of the deliverable cluster resource hours will be provided to Lab and campus collaborations under the program, and NERSC will provide archival storage (an aggregate of five terabytes per year) to users of the Lab cluster. NERSC will also provide staff to configure and administer the system.

The cluster will consist of 160 Intel Pentium III processors in 80 nodes, with one gigabyte of main memory in each node. Each node will be connected by a high-performance interconnect from Myrinet. The cluster will have a total of 500 gigabytes of shared disk space

“Atlhough they are still a maturing technology, clusters today can match the performance of high-end supercomputers from just a few years ago,” said Tammy Welcome, head of NERSC’s cluster computer project. “Buying a cluster now allows us to study a possible technology path of the future as well as provide limited computational support to several research efforts here at the Lab.”

NERSC already has experience in configuring and operating clusters. The Parallel Distributed Systems Facility (PDSF) is a 264-processor networked, distributed computing environment used to meet the detector simulation and data analysis requirements of large scale High Energy Physics and Nuclear Science investigations. NERSC’s Future Technologies Group has also operated 12-node and a 32-node research clusters and develops software to improve the performance of Linux-based clusters.