English (United Kingdom)

https://curated-unify.zendy.io/wp-json/zendy-region/v1/featured_content/oa?rat=en

https://curated-unify.zendy.io/wp-json/zendy-region/v1/highlighted_journal/

Global: Zendy Plus annual

Presents the access of premium content as premium feature

Premium Content

Presents the keyphrase highlighting as premium feature

Keyphrase Highlighting

Presents the summarisation as premium feature

Summarisation

Insights

Presents the pdf analysis as premium feature

PDF Analysis

Presents the zaia usage as premium feature

ZAIA

Global: Zendy Tools annual

Global: Zendy Free Plan

Global: Zendy Tools monthly

Global: Zendy Plus monthly

MPI+threads is gaining prominence as an alternative to the traditional MPIeverywhere model in order to better handle the disproportionate increase in thenumber of cores compared with other on-node resources. However, thecommunication performance of MPI+threads can be 100x slower than that of MPIeverywhere. Both MPI users and developers are to blame for this slowdown.Typically, MPI users do not expose logical communication parallelism.Consequently, MPI libraries use conservative approaches, such as a globalcritical section, to maintain MPI's ordering constraints for MPI+threads, thusserializing access to parallel network resources and hurting performance.  To enhance MP+threads' communication performance, researchers have proposedMPI Endpoints as a user-visible extension to MPI-3.1. MPI Endpoints allows asingle process to create multiple MPI ranks within a communicator. This couldallow each thread to have a dedicated communication path to the network andimprove performance. The onus of mapping threads to endpoints, however, wouldthen be on domain scientists. We play the role of devil's advocate and questionthe need for user-visible endpoints. We certainly agree that dedicatedcommunication channels are critical. To what extent, however, can we hide thesechannels inside the MPI library without modifying the MPI standard and thusunburden the user? More important, what functionality would we lose throughsuch abstraction? This paper answers these questions through a new MPI-3.1implementation that uses virtual communication interfaces (VCIs). VCIs abstractunderlying network contexts. When users expose parallelism through existing MPImechanisms, the MPI library maps that parallelism to the VCIs, relieving domainscientists from endpoints. We identify cases where VCIs perform as well asuser-visible endpoints, as well as cases where such abstraction hurtsperformance.

How I Learned to Stop Worrying About User-Visible Endpoints and Love MPI