Empirical Comparisons of Some Design Replay Algorithms

Abstract

Although most design replay techniques have been empirically tested against some performance program, there has been very little empirical evidence published that compares various approaches on the same problems to determine the source of power. Six different design replay algorithms based on approaches in the literature are implemented and tested on 20 different design replay problems. The resulting data indicate that there is a trade-off between efficiency and autonomy for certain types of adaptation strategies. Based on some of the lessons drawn from this data, a new algorithm, REMAID, has been developed. This algorithm recognizes two different types of mis-matches between previous experience and current problems: detours and pretours. The REMAID strategy takes advantage of its knowledge of mis-matches to improve replay autonomy without sacrificing efficiency. The success of the REMAID algorithm is empirically verified.