Emulation facility for data-driven real-time multi-processing

Abstract

Minimizing overhead resulting from context switching is a prerequisite for effective real-time multi-processing. Our heterogeneous data-driven processor has shown its multi-processing capability without extending turn-around time of any process as long as enougth processing resource is available. However, the heterogeneous multi-processor system connected by a non-uniform network can yield bottlenecks that prevent it from meeting performance requirements unless program allocation is optimized during preprocessing. To avoid overload at the bottlenecks, we need a prototyping scheme to verify each processor's load. This paper discusses an emulation facility for an elastic pipeline structure. To effective emulate our data-driven processor, fine-grained parallel processing is required because the elastic pipeline structure makes up an autonomous processor without a centralized clock. By testing our data-driven processor used as an emulation facility to emulate fine-grained pipeline parallelism, we show its effectiveness.