Parallel Programming Models for the Extreme Scale Era
TimeMonday, November 12th1:30pm - 1:52pm
DescriptionParallel programming has always been challenging. In addition to the traditional challenges of (sequential) programming, parallel programming has to deal with complex issues of performance such as load imbalances, long critical paths, and communication overheads, as well as correctness issues arising from race conditions. But with the current and upcoming era in parallel computing the challenges have further increased. There are hardware issues of heterogeneity as well as static and dynamic performance variability, and application-induced issues of multi-module multi-physics programming, adaptive refinements, and strong-scaling. I will explain the programming challenges these issues create, and describe some programming concepts and abstractions that are useful in dealing with them. I will also briefly discuss some of the non-traditional parallel programming approaches, in addition to the dominant MPI-OpenMP paradigm. The objective of this brief talk will be to convey some of the excitement of this research area that I am passionate about, and to provide pointers to research that you may follow later.