Phase 1 of NERSC’s IBM RS/6000 SP system, with 512 processors and a peak performance of 410 gigaflop/s, will be replaced by the 3 teraflop/s Phase 2 system early in 2001.

When Phase 1 of NERSC’s new IBM RS/6000 SP computer system went into full service on April 5, 2000, the launch was 25 days ahead of schedule. As a result of getting the system up and running sooner than expected, NERSC was able to provide users with one million more MPP hours than anticipated for the FY 1999 allocation period.

This achievement, which capped a very complex procurement process, was the result of a five-month effort by a team of 30 employees (listed below) led by High Performance Computing Department head Bill Kramer. They installed, configured, and customized the SP system for the NERSC production environment and users, worked with early users to test and refine the system, and created training programs and documentation as well as demonstration applications. Although the team had little prior experience with SP systems and there were no permanent IBM staff on site yet, they learned quickly on their own and created innovative solutions to the problems they encountered.

In keeping with this pioneering spirit, NERSC was the first site to commit all its files to IBM’s new General Parallel File System (GPFS). Our SP team helped IBM find and resolve issues that will benefit all of our users and other sites as well. NERSC’s success with GPFS has already prompted other sites to adopt it. But what’s most important is that NERSC users are keeping the SP constantly busy and are happy with the system’s performance.