Speaker: Girija Narlikar, CMU
Making Parallel Applications Run Faster on DSM SystemsDate: November 30, 1995
Abstract: Advances in processor and network technology make it increasingly desirable to use a workstation cluster as a multicomputing platform for parallel applications. However, run time support is still needed to hide latency and increase concurrency in order to get good performance without increasing the load on the programmer. This talk describes the work I did this summer at DEC SRC, on exploring the use of run time optimizations to speed up parallel applications running on a cluster of Alphas. I will describe some general optimizations that make use of data-access and execution patterns of parallel applications written in a distributed shared memory system.