Competitive execution: A method to exploit idle workstations

Abstract

This paper presents competition protocols to speed up distributed programs on a network of shared work-stations in the background by exploiting their wasted computing capacity, without interfering with foreground processes. Competition protocols are transparent operating system facilities that involve creating multiple instances (called clones) p(1), p(2), etc. of a process P on different processors, and making clones "compete", i.e., attempting to guarantee on a moment-by-moment basis that the output of the clone that is farthest "ahead" is fed to the rest of the computation, and that the entire application's performance tracks that of the clone which is farthest ahead. We show that competition protocols speed up distributed programs, especially with smaller granularity, We also show that competition protocol offer more speedup far distributed programs than for sequential programs.