QUANTIFYING THE PERFORMANCE DIFFERENCES BETWEEN PVM AND TREADMARKS

This paper compares two systems for parallel programming on networks o f workstations: Parallel Virtual Machine (PVM), a message-passing syst em, and TreadMarks, a software distributed shared-memory (DSM) system, The eight applications used in this comparison are Water and Barnes-H ut from the SPLASH benchmark suite; 3-D FFT, Integer Sort (IS), and Em barrassingly Parallel (EP) from the NAS benchmarks; ILINK, a widely us ed genetic linkage analysis program; and Successive Over-Relaxation (S OR) and Traveling Salesman (TSP), Two different input data sets are us ed for five of the applications, We use two execution environments, Th e first is a 155 Mbps ATM network with eight Spare-20 model 61 worksta tions; the second is an eight-processor IBM SP/2. The differences in s peedup between TreadMarks and PVM depend mostly on the applications, a nd only to a much lesser extent on the platform and the data set used, In particular, the TreadMarks speedup for six of the eight applicatio ns is within 15% of that achieved with PVM, For one application, the d ifference in speedup is between 15% and 30%, and for another, the diff erence is around 50%, We identified four important factors that contri bute to the lower performance of TreadMarks: (1) extra messages due to the separation of synchronization and data transfer, (2) extra messag es to handle access misses caused by the use of an invalidate protocol , (3) false sharing, and (4) diff accumulation for migratory data, We have quantified the effects of the last three factors by measuring the performance gain when each is eliminated, Of the three factors, Tread Marks' use of a separate request message per page of data accessed is the most important. The effect of false sharing is comparatively low R educing diff accumulation benefits migratory data only when the diffs completely overlap, When these performance impediments are removed, al l of the TreadMarks programs perform within 25% of PVM, and for six ou t of eight experiments, TreadMarks is less than 5% slower than PVM. (C ) 1997 Academic Press.