MCS 572 Individual PSC TCS Cluster Starter Problem
Spring 2003
Professor F. B. Hanson
DUE Monday 10 March 2003 in class (this is individual,
not group, homework, so it must be your own work)
Sources:
Measure the speed-up and efficiency on the PSC TCS prototype Cluster using
one (1) the codes
Specifications:
- Run p = 1, 2, 4, 8 and 16 processor jobs;
- Use modifications where necessary of the QSUB MPI job scripts
from the class
MPI-Examples Page such as
noting these are for 1, 4 and 8 processor jobs only;
- Test and Compare the optimization options
*
-O0, -O2 (= -O), -fast, -O5 for C
OR
*
-O0, -O2 -O3, -O5 -fast for F90.
Are they worth it for the simple
examples. Are -fast for C OR -O5 -fast for F90 the best
performing as claimed by the
handed out in class?
- Take at least 2 combinations of processors and optimization options
and run a total sample 3 runs to get an estimate of the timing variability
of the MPI wall timer and compute the average.
- Code modifications are permissible if documented by comments within
the code, especially if effort if made to increase the workload or the
parallel efficiency of the
toy codes for PI by the midpoint rule and the TRAPEZOIDAL rule.
For instance,
*
for mpi_trap.* replace the integrand
5*x4 by a more computationally demanding function,
inline the function procedures/subroutines, use a larger number of subintervals
n, or declare n globally by #define/parameter
statements,
OR
*
for mpi_pi.* inline the scanf/read statements
as for/do loops using lots of the larger subintervals, find out
how to use quad precision while changing the error reference value to more
than 25 digits for pi or change the midpoint rule to a more
computationally intensive rule like Simpson's rule.
Summary Report:
Prepare
- Summary Page with tables of
- Speed-ups and efficiencies of each processors-options case
(use average time if relevant).
- Total wall times results for each processors-options case.
- For the 2 wall time variability cases, report the wall times
list the triple set, the average and the standard deviation.
- Documentation:
- Listing of the code, with documentary comments, used.
- Listing of the job script used.
- Formatted Output Results for codes.
- List with comments what modifications were made in the code.
Be sure to label all above items for identification.
You must correct the code to remove error messages, especially, and
warning message if possible (job script warning about the terminal
is standard and is not necessary to remove it.
Try to make the code fit the parallel machine model.
Notes:
Please report to Professor Hanson any problems by email.
Web Source:http://www.math.uic.edu/~hanson/tcs03start.html