Analysis of Scheduling Algorithms for Parallel Computers with Virtual Shared Memory

The growing presence of massively-parallel systems using physically distributed memory has promoted the development of new programming models for such systems. An example of such a system offering a global address space to simplify programming is the Shared Virtual Memory system. However, scheduling inside and outside of a program has to be modified to allow such systems to run at an acceptable level of performance. After giving an introduction into SVM and scheduling issues on parallel computers, this article describes the implementation of a simulation system that will allow to test and compare different scheduling algorithms on a simulated parallel computer using SVM as its primary programming model. At the end, a sample session with the analyzation tool PARvis is described