Using Analytical Performance Prediction for Architectural Scaling
Mark Clement, Michael J. Quinn · 1994
Recent advances in the power of parallel computers have made them attractive for solving large computational problems. High speed parallel processors ranging from clusters of workstations to Massively Parallel Processing (MPP) machines are becoming increasingly available to the scientific community [4]. In this paper, the term "scalability" is used to indicate how well an application is able to utilize a varying number of processors. This research develops an analytical model which is used to predict the scalability of data-parallel programs as the parallel architecture varies. This model can be used to determine the portability of applications as well as the impact of architectural features on performance. 1 Introduction One of the most important advances in high performance computing is the increasing availability of commercial multicomputers. For our purposes, multicomputers will be defined as parallel computers where the processors do not have shared memory [22]. This definition ...