Many MIMD message-passing multiprocessors are large-scale, but the programming environment based on message-passing and the limited memory on each node preclude many applications. Shared-memory multiprocessors do not have these restrictions, but it is very difficult to build large-scale ones. This research is to investigate the methods of building shared virtual memory systems on large-scale multiprocessors without shared memory, allowing parallel programs based on shared- memory model to run on these large-scale multiprocessors. The research provides the framework upon which to implement systems and programming languages based on shared-memory model that effectively harness the resources of large-scale stock multiprocessors.
