Roadmap: MPI One-Sided Communication
One-sided communication methods were added to MPI as a part of the MPI-2 improvements and were greatly expanded in MPI-3 by including support for shared memory windows, windows with dynamically attached memory, request-based communication calls, and more window locking mechanisms. On Stampede2 and Frontera, the one-sided communication methods implemented in the Intel MPI and MVAPICH2 libraries use the Remote Direct Memory Access (RMA or RDMA) functionality provided by low-latency interconnect fabrics such as Omni-Path and InfiniBand. In this roadmap, we will introduce the various components of MPI RMA and how to use them.
All exercises and examples are verified to work on Stampede2 and Frontera.
This is the fifth of five related roadmaps in the Cornell Virtual Workshop that cover MPI. To see the other roadmaps available, please visit the complete roadmaps list.
Objectives
After you complete this roadmap, you should be able to:
- Identify basic concepts of one-sided communication in MPI programming
- Define the term one-sided communication
- Explain how RMA improves data transfer events
- Identify the three RMA communication calls supported by MPI
- Define the target_rank and target_datatype identifiers
- Demonstrate synchronizing MPI processes
- List important considerations when using RMA calls
- Demonstrate use of dynamically allocated memory
- Explain the need for creating windows with shared memory
Prerequisites
- A basic knowledge of parallel programming and MPI. Information on these prerequisites can be found in other topics (Parallel Programming Concepts and High-Performance Computing, MPI Basics).
- Ability to program in a high-level language such as Fortran or C.
- MPI Collective Communications logically precedes this roadmap but is not a prerequisite.
Requirements
System requirements include:
- A TACC to login to Stampede2 or Frontera
- A computation allocation for Stampede2 or Frontera