Not only must the aggregate capability be enhanced, but special attention must given to improving the component subsystems of each class of capability resource (PVPs, MPPs, SMPs); e.g., fast CPUs, increased memory capacity and memory bandwidth, increased interprocess communication (mechanisms for both shared and distributed memory programming models) and disk I/O performance. It is clear that improving these component subsystems will require next generation technology in both hardware and software.
The next system at NERSC will co-exist with the current T3E MPP and the J-90 PVP cluster (with upgrades). Access to all classes of production capability resources (PVPs, MPPs, SMP systems and clusters, c.f. section 4) is essential. Access to the many classes of computers has the benefit of allowing the user to appropriately match the resource capability to the scientific task or job. The current resources meet one HPCCI goal of exporting parallel computing technology to a broad user base. An additional benefit of the co-located resources at NERSC is that it allows to set up test-beds for any distributed computing environment in a production facility.