Message Passing Interface; MPI
- A standardized and portable message-passing system
- Not a new language —— Library of functions
- Scales to over 100k processors
- Runs on distributed memory cluster systems.
- Portable
- Explicit parallel programming (unlike OpenMP)
- No shared variables (unlike OpenMP)
MPI’s Execution model
- All processors execute same program, but with difference data. (no fork & join)
- Variable on each are all private.
- Data that must be shared is explicitly send between processors.
Distributed Memory Cluster
Data sent over Network:
- Ethernet
- 10M~1GBit/Sec
- InfiniBand
- 100GBit~
- MPI on SMP machine
- You can run MPI program on SMP (Symmetric Multiprocessing) machine.
- Variables are not shared (all private).
Software Distribution
Brief Tutorials for Programming Languages
Last modified
5 years ago
Last modified on 04/27/20 17:10:12
Attachments (2)
- MPIexecModel.png (24.9 KB ) - added by 9 years ago.
- CypressNodes.png (25.3 KB ) - added by 9 years ago.
Download all attachments as: .zip
Note:
See TracWiki
for help on using the wiki.