2004/03/11 by George Candea
Computer Science · #cs.OS #cs.DC
published as WIP Session, USENIX Technical Conference, Boston, MA, June 2004 · 6 pages; submitted to 11th ACM SIGOPS European Workshop
arxiv created 2004/03/11 · arxiv updated 2009/12/01
Many dependability techniques expect certain behaviors from the underlying subsystems and fail in chaotic ways if these expectations are not met. Under expected circumstances, however, software tends to work quite well. This paper suggests that, instead of fixing elusive bugs or rewriting software, we improve the predictability of conditions faced by our programs. This approach might be a cheaper and faster way to improve dependability of software. After identifying some of the common triggers of unpredictability, the paper describes three engineering principles that hold promise in combating unpredictability, suggests a way to benchmark predictability, and outlines a brief research agenda.