Introduction to Parallel Computing

1
Introduction to Parallel Computing

• Dr. Guy Zimmerman
• Bowling Green State University
• Bowling Green, Ohio
• gzimmer_at_cs.bgsu.edu
• www.cs.bgsu.edu/zimmerman

2
Outline

• About BGSU
• Course Overview
• Course Objectives I, II, III
• Hardware/Software
• Projects
• Impressions

3
About BGSU

• 350 Undergraduate majors
• 40 Graduate Students
• CS in Arts Sciences

4
Course Overview

• Focus Message Passing , NOWs
• Text Parallel Programming, Wilkinson Allen
  (http//www.coe.uncc.edu/abw/parallel/par_prog/)
• Audience/Prerequisite

5
Objectives - I

• Flynns Taxonomy
• Languages for parallel computing, including PVM,
  OpenMP and Pthreads
• The Message passing paradigm
• Parallel Processing speedup issues, including
  Amdahls and Gustafasons Laws
• Decomposition methodologies for parallel program
  development

6
Objectives - II

• Load balancing issues for parallel programs
• Parallel Architectures and effect of architecture
  on design/implementation of a parallel algorithm
• Quantifying Speedup empirical tools and
  theoretical models
• Current issues in parallel processing

7
Objectives - III

• Design, implement, test and debug an MP parallel
  application
• Design, implement, test and debug a parallel
  application program using OpenMP
• Parallelize an existing application using an
  appropriate parallel programming paradigm
• Explain the tradeoffs that result from using a
  specific programming paradigm for a given problem
  class.

8
Hardware

• 2 Processor DEC ALPHA
• 4 Processor SGI Power Challenge
• SGI 02 workstations
• Ohio SuperComputer Center
• SGI ORIGIN, 32 Processor

9
Software C libraries

• PVM ( Parallel Virtual Machine )
• OpenMP
• pthread
• What about FORTRAN?

10
Projects - I Computing the Mandelbrot Set

• Easy, but interesting
• Master/Slave example
• Can include a visual component
• Decomposition issues
• Timing issues

zk1 zk 2 c
11
Projects - II Solving the Heat equation

• Decomposition and Synchronization issues
• Numerical issues
• practical application

12
Projects - III Parallelizing using OpenMP

• Use OMP directives to parallelize
• matrix mult, mandelbrot and 2d heat
• Intro to OpenMP
• Scheduling, dependency issues
• Easy (end of semester project)

13
Impressions

• Students loved it
• A lot of work
• Prerequisites?
• PVM?
• Other