Queue programs characterization using performance bounds

Abstract

The special characteristics and simplicity of queue computing offer an attractive model for building state of the art processors with low hardware complexity, low power consumption, small memory footprint, and high parallelism. This paper presents a performance bounds characterization of the queue computation model. Performance bounds gives an approximation of the upper limit performance that can be expected from a given system for a given application. Thus, this characterization provides useful information about the strengths of a queue-based computer system. This design space exploration approach is useful to estimate the performance of a computer system without having to build the actual hardware. We present a performance bounds model based on the Lower Bound Execution Time (LBET) that has been implemented in the queue compiler. Our infrastructure profiles applications and gathers basic block execution statistics to feed and solve the LBET equation to approximate its performance. Our results suggest that a queue processor is capable of achieving similar performance to a conventional high-performance superscalar register machine.