Home
About
Services
Research
News
Events
Contact

   
Virtual Workshop Modules
Collective Communication I
Introduction
1. Overview
2. MPI Routines
   2.1 Characteristics
   2.2 Barrier Routine
   2.3 Data Movement
   2.3.1 Broadcast
   2.3.2 Gather and Scatter
   2.3.3 Gatherv and Scatterv
   2.3.4 Allgather
   2.3.5 All to All
   2.4 Global Computing
3. Performance Issues
   3.1 Broadcast
   3.2 Scatter
References
Lab Exercise 1
Lab Exercise 2
Lab Exercise 3
Quiz
Evaluation
Questions?
3. Performance Issues
  • A great deal of hidden communication takes place with collective communication
  • Performance depends greatly on the particular implementation of MPI
  • Because there may be forced synchronization, not always best to use collective communication

    • © 2008 Cornell University | Cornell University Center for Advanced Computing | Copyright Statement