Parallel Solution of 2-D Heat Equation Using Laplace Finite Difference
-
Upload
jada-floyd -
Category
Documents
-
view
32 -
download
1
description
Transcript of Parallel Solution of 2-D Heat Equation Using Laplace Finite Difference
OAK RIDGE NATIONAL LABORATORYU.S. DEPARTMENT OF ENERGY
Parallel Solution of 2-D Heat Equation Using Laplace Finite Difference
Presented by Valerie SpencerMentored by Jim Kohl
Oak Ridge National Laboratory RAM Internship
June 03 – August 16, 2002
This research was performed under the Research Alliance for Minorities Program administered through the Computer Science and Mathematics Division, Oak Ridge
National Laboratory. This Program is sponsored by the Mathematical, Information, and Computational Sciences Division; Office of Advanced Scientific Computing
Research; U.S. Department of Energy. Oak Ridge National Laboratory is managed by UT-Battelle, LLC, for the U.S. Department of Energy under contract DE-AC05-
00OR22725. This work has been authored by a contractor of the U.S. Government under contract DE-AC05-00OR22725. Accordingly, the U.S. Government retains a
nonexclusive, royalty-free license to publish or reproduce the published form of this contribution, or allow others to do so, for U.S. Government purposes.
OAK RIDGE NATIONAL LABORATORYU.S. DEPARTMENT OF ENERGY
Table of Contents Purpose
Background work
Problem Statement
Linux
Parallel Virtual Machine
Two parts
Using pvm
Two models
Parallel Program Used
Jacobi Iteration Method
Parallel cluster used
Results
Conclusion
OAK RIDGE NATIONAL LABORATORYU.S. DEPARTMENT OF ENERGY
Purpose
The research involves solving a multi-dimensional conduction heat
transfer problem on a Linux cluster using Parallel Virtual
Machine (PVM). Temperature distribution in a X*Y size wall can
be determined by solving the Laplace Equation. The larger the
size gets, the more computing power is needed.
OAK RIDGE NATIONAL LABORATORYU.S. DEPARTMENT OF ENERGY
Background Work
lab experiment at Alabama A&M University - determine the
temperature distribution inside a medium of size 1.0 m x 1.0 m
using the finite difference method
heat equation (2-D array notation Laplace Eq.)
[(∂2T / ∂x2)+(∂2T / ∂y2)]I,J = 0
T(I,J) = (TE + TW + TN + TS) / 4, where
TE = T(I+1,J)
TW = T(I-1,J)
TN = T(I,J+1)
TS = T(I-1,J)
OAK RIDGE NATIONAL LABORATORYU.S. DEPARTMENT OF ENERGY
Background Cont’d
Fortran program -
Microsoft Developer
Studio software
Visual image of the
temperature
distribution -Tecplot
Software
X
Y
0 0.5 10
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1T
328.041
316.216
304.392
292.568
280.743
268.919
257.095
245.27
233.446
221.622
209.797
197.973
186.149
174.324
162.5
Temperature Distribution
OAK RIDGE NATIONAL LABORATORYU.S. DEPARTMENT OF ENERGY
Problem Statement
Decrease experiment cycle time for solving the heat equation
over large problems
Solution:
Parallelize the heat equation simulation program and run using
multiple computers in a cluster
OAK RIDGE NATIONAL LABORATORYU.S. DEPARTMENT OF ENERGY
a sophisticated multitasking virtual memory operating system
directly controls the hardware
provides:
true multitasking
virtual memory
shared libraries
demand loading
shared copy-on-write executables
TCP/IP networking
file systems
OAK RIDGE NATIONAL LABORATORYU.S. DEPARTMENT OF ENERGY
software that permits a heterogeneous collection of Unix and/or
NT computers connected together by a network to be used as a
single large parallel computer
OAK RIDGE NATIONAL LABORATORYU.S. DEPARTMENT OF ENERGY
Two Part PVM System
daemon is a special purpose process that runs on behalf of the
system to handle all the incoming and outgoing messages; it is
represented by “pvmd3” or “pvmd” and any user with a valid
login id can install and execute this on a machine
library of routines that allows the computers to interact “in
parallel” (resource and task management “addhost” and
“spawn”, pack and send messages…)
OAK RIDGE NATIONAL LABORATORYU.S. DEPARTMENT OF ENERGY
Using PVM
The problem must be able to be broken down into several tasks
functional parallelism - breaking the application into different
tasks that perform different functions
data parallelism - having several similar tasks that are the
same, solve over different parts of the data
OAK RIDGE NATIONAL LABORATORYU.S. DEPARTMENT OF ENERGY
Two Models of PVM Codes
master/worker model - the master task creates all other tasks
that are designed to work on the problem, then coordinates the
input of initial data to each task, and collects the output of
results from each task
hostless model - the initial task spawns off copies of itself as
tasks and then starts working on its portion of the problem while
the created tasks immediately begin working on their portion
OAK RIDGE NATIONAL LABORATORYU.S. DEPARTMENT OF ENERGY
Jacobi Iteration Method
Based on solving for every
variable locally with respect
to the other variables
One iteration of the method
corresponds to solving for
every variable once
Simple parallel data structure
Processes exchange
columns with neighbors Process 0 Process 1 Process 2 Process 3
Boundary Point
Interior Point
OAK RIDGE NATIONAL LABORATORYU.S. DEPARTMENT OF ENERGY
TORC(Tennessee Oak Ridge Cluster)
Consists of 18 processors
1 Dell PowerEdge 1300
Dual PIII 450MHz, 512MB RAM
4 Dell PowerEdge 6350
Quad PII Xeon 450MHz, 2GB
RAM(node4 – 1GB RAM)
PGI v3.2 components
pgf77 – fortran 77 compiler
pgf90 – fortran 90 compiler
pghpf – high performance
fortran compiler
OAK RIDGE NATIONAL LABORATORYU.S. DEPARTMENT OF ENERGY
ResultsTest Run #1
Additional Processor to Speed Ratio vs. Ideal Speedup for Different Medium Sizes Ranging from (1 x 1)m to (10 x 10)m
0
2
4
6
8
10
12
14
16
18
20
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18
Number of Processors
Speedup
Ideal
100 x 100
200 x 200
500 x 500
1000 x 1000
OAK RIDGE NATIONAL LABORATORYU.S. DEPARTMENT OF ENERGY
ResultsTest Run #2
Additional Processor to Speed Ratio vs. Ideal Speedup for Different Medium Sizes Ranging from (1 x 1)m to (10 x 10)m
0
2
4
6
8
10
12
14
16
18
20
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18
Number of Processors
Speedup
Ideal
100 x 100
200 x 200
500 x 500
1000 x 1000
OAK RIDGE NATIONAL LABORATORYU.S. DEPARTMENT OF ENERGY
Conclusion
Larger problem, more computational power needed
Clustered personal computers provides adequate computing
power
Parallelization is not needed for small problems
PVM on a Linux system is an excellent tool to solve large
engineering problems
OAK RIDGE NATIONAL LABORATORYU.S. DEPARTMENT OF ENERGY
About Myself
Alabama A&M University, Senior
Mechanical Engineering, G.P.A – 3.8
B.S. - May 2003
U.S. Navy- Nupoc Program
OAK RIDGE NATIONAL LABORATORYU.S. DEPARTMENT OF ENERGY
Acknowledgements
I appreciate Dr. Z. T. Deng’s decision of selecting me to be a
part of the Summer 2002 RAM Program. I appreciate Dr. Jim
Kohl for guiding my research, dedicating time to discuss my
findings, and facilitating resources from the Computer
Science and Mathematics Division. I would also like to
acknowledge Debbie McCoy for organizing the RAM
Program that allowed me to spend the summer gaining
significant information at ORNL. Other acknowledgements:
Stephen Scott and Cheryl Hamby.