Large scale production projects require long-time access to either a top-end high performance computer system or a large cluster of high performance systems. An individual computation will be of the scale that can be executed on current high-end shared-memory parallel machines (memory requirement normally below 10 GB) or on a relatively small segment of an MPP, and the main requirement on the hardware configuration is that it provides fast throughput. The access need to this system may be intermittent or constant.
Medium scale projects have been the bread-and-butter of NERSC's operation; with the current spread of high-end workstations and small SMPs in the laboratories we would classify as medium-scale those projects that require extensive use of a small multiprocessor; many ER scientists would not now have good access to such a machine and many that do have access find their productivity enhanced if they can use, intermittently, a high-end computational facility such as the NERSC C90 or its expected Origin-2000.
Small scale projects may utilize the resources of the High Performance Computing Access Center for a variety of reasons: the required hardware or software is not available locally, the center is convenient for remote collaborations, the center's consulting facilities or charging policies make it attractive. These projects can provide the fill for gaps in effective management of larger scale jobs.