Parallel%20Apps

1
Parallel Apps
November 6, 2000

• Hyang-Ah Kim
• Brenda Liu
• SoYoung Park

2
Outline

• Introduction
• Barnes background
• Barnes optimizations
• Ocean background
• Ocean optimizations
• Conclusion

3
Introduction

• Minimum problem size
• Scale application performance
• Programming models

• Parallel efficiency?
• (speedup over uniprocessor) / p

4
Barnes Background

• N-body galaxy simulation

• Communication pattern?
• Irregular
• Hierarchical

5
Barnes Problem Size

• Optimizations visited
• Data placement
• Dynamic partitioning
• Prefetching
• Work needed to scale is algorithmic

6
Scaling Performance

• Performance change from 32 to 128 processors?
• Degradation Communication-computation ratio,
  communication pattern, load balance, locality,
  synchronization
• How can they be overcome?
• Increase problem size
• Application restructuring

7
General Findings

• Scaling to 128 processors without any change

8
Scaling Barnes

• Memory bottleneck building shared tree (31 in
  128-proc vs. 2 is uniprocessor)

• Original algorithm globally shared tree

9
Scaling Barnes
10
Scaling Barnes

• New algorithm MergeTree

11
Ocean Background

• Ocean simulation using multigrid solver

• Communication pattern?
• Nearest neighbor iterative
• Hierarchical

12
Ocean Problem Size

• Optimizations visited
• Processor-centric array data structures
• Data placement
• Prefetching
• Work needed to scale is difficult

13
Programming Models

• Options
• Shared Address Space
• Message Passing
• SHMEM
• Motivation
• if application is regular / predictable?
• If we can use similar algorithms and partitions
  across the models?

14
Ocean Discussions
15
Ocean Discussions
16
Conclusion

• Some guidelines
• Load balancing for moderate systems,
  communication for large systems
• Data partition placement
• Very application dependent
• Optimization
• Programming model