題目：MPI Communication Challenges in Irregular Data-Driven Applications
講師：Dr. Pavan Balaji, Argonne National Laboratory, USA
概要：Modern applications are increasingly moving towards an irregular data-driven computational model where how they compute and which processes they communicate with fundamentally relies on the data being processed. This results in frequent communication of small data elements with a large number of processes, ability to process such messages asynchronously and on-demand, notions of latency-hiding by leveraging very deep queues of outstanding operations and the capability to perform efficient communication when hundreds of threads try to communicate simultaneously. In this talk, I’ll first discuss some of the large computational applications at Argonne and their computational trends. Then I’ll discuss some of the optimizations that we are exploring within the MPI library to address these challenges, while looking forward to where these applications are planning to evolve to in the next few years.