RPM - A RAPID PROTOTYPING ENGINE FOR MULTIPROCESSOR SYSTEMS

In multiprocessor systems, processing nodes contain a processor, some cache memory, and a share of the system memory, and connect through a scalable interconnection. The system memory partitions may be shared o r disjoint (message passing). Within each class of systems, many archi tectural variations are possible. Fair comparisons among systems are d ifficult without a common hardware platform to implement the different architectures. RPM (Rapid Prototyping engine for Multiprocessors), a hardware emulator for the rapid prototyping of various multiprocessor architectures, provides this platform. The authors describe its archit ecture, performance, and prototyping methodology. Reprogrammable contr ollers implemented with field-programmable gate arrays emulate the tar get machine's hardware. The processors, memories, and interconnections are off the shelf, and their relative speeds can be modified to emula te various component technologies. The authors also compare RPM with o ther rapid prototyping approaches. Because emulation is orders-of-magn itude faster than simulation, an emulator can run problems with large data sets more representative of the workloads for which the target ma chine is designed. An emulator can also accomplish more reliable perfo rmance evaluation and design. Finally, because an emulator is a real c omputer with its own I/O, every emulation is an actual incarnation of the target and can run several different workloads.