Write detection in home-based software DSMs

Write detection is essential in multiple writer protocols to identify writes to shared pages so that these writes can be correctly propagated. This paper studies different write detection schemes in a home-based software DSM system called JIAJIA. It compares the performance of three write detection schemes: the traditional virtual memory page fault write detection scheme which write-protects both home and cached pages at the beginning of an interval, a cache only write detection scheme which does not detect writes to home pages but invalidates all cached pages at the beginning of an interval, and an API write detection scheme which requires the programmer or pre-complier to explicitly records writes in program. Evaluation with some well-known DSM benchmarks reveals that tradeoffs of different write detection schemes vary with data (home) distribution and memory reference patterns of applications.