Measuring Setl Performance

Candidate: Shields,Lynwood David


Current computer technology is being driven by the hardware advances that have provided a constant and dramatic decrease in the cost of elementary hardware operations. This has made more feasible the use of high-level languages that permit program development without the constant attention to detail needed to achieve efficient execution that characterizes low-level languages; indeed, such languages can be realized by a combination of microcode and special-purpose VLSI chips. However, effective use of this technology requires an understanding of the underlying performance issues. We have analyzed the problem of measuring performance of high-level languages by studying in detail one such language, SETL, and have developed a set of measurement tools addressed both to the user and the implementor. Our thesis is that such measurement efforts must aim to provide measurement tools that can be integrated into the system, but only after their efficacy has been demonstrated by their use on real programs. This work has resulted in prototype versions of four program profilers, each providing a specific view of SETL performance; we discuss their use in analyzing, and then improving, the performance of actual SETL programs. We also discuss the implementation of the hard code system that provides an essential starting point for evaluating the effectiveness of the representation sublanguage provided by SETL. Finally, we indicate some ways in which SETL performance can be improved.