Partitioned Collective Communication
Event Type
Workshop
Online Only
Extreme Scale Comptuing
Parallel Programming Languages and Models
Performance
W
TimeSunday, 14 November 202111am - 11:30am CST
LocationOnline
DescriptionPartitioned point-to-point communication and persistent collective communication were recently standardized in MPI-4.0. They both offer performance and scalability advantages over MPI-3.1-based communication when planned transfers are possible in an MPI application. Partitioned collectives will be particularly useful for multithreaded, accelerator-offloaded, and/or hardware-collective-enhanced MPI implementations driving suitable applications, as well as for pipelined collective communication (e.g., partitioned allreduce) with single consumers and producers per MPI process. These operations also provide load imbalance mitigation. Halo-exchange codes arising from regular and irregular grid/mesh applications are a key candidate class of applications for this functionality. For such codes, efficient support for double-buffering is a key driver for partitioned collective operations. The option of providing local and incomplete modes for initialization procedures is mentioned (which could also apply to persistent collective operations); these semantics interact with the Pbuf_prepare concept and the progress rule.
Author/Presenters
