Optimizing Parallel Programming with MPI

1
Optimizing Parallel Programming with MPI
Computer Systems Research Lab 2007-2008 Michael
Chen Period 3
2
MPI

• Message passing interface
• C, C, and Fortran
• De facto standard
• Updated with MPI libraries

3
Game of Life

• Cellular automaton (model), created by Conway
• Models social activity
• Patterns
• Why use it?
• Alternatives

4
(No Transcript)
5
Methodology Problems

• From 2nd quarter, I had a basic game of life
  program
• Optimizing efficiency
• Question latency vs. computing power
• Organization wasnt optimal, so code-tweaking was
  needed for
• Display method
• Static variables
• Testing program
• File i/o strings

6
Results

• Ran eleven different scenarios with multiple
  trials
• Results not quite as expected increased run-time
  with more processors!
• But, general trend for improved efficiency with
  more rows/columns around 4

7
Results (cont.)?
8
Possible explanations

• Overhead in passing information
• Unnecessary code
• Larger role of latency

9
Future possibilities

• The number 4?
• Reducing overhead
• More processors
• Heat map
• More testing