Date: Wed, 30 Oct 2002 13:35:42 -0800 (PST) From: kris stewart To: cs575@rohan.sdsu.edu Subject: Postponement of Report2 Deadline (by one week) Hello CS 575 Students, Postpone: Report 2 deadline to 6 Nov 02 I have been working with the Sys Admin for Rohan (Ron Nash) and we have discovered a consistent and reportable (to Sun Microsystems) problem with the gethrtime routine that I encouraged you to explore for this second Computational Experiment. Since this is an important instance of a platform-dependent performance issue, I hope you will enjoy working through this with me. I was able to produce data for the cases M K N 10 10 100 10 20 200 10 40 400 10 80 800 negative value for sgeco 10 160 1600 10 320 3200 we have tried also using gethrvtime which yielded even more negative values. Since the man page for gethrtime/gethrvtime clearly indicates that this high resolution system timer are guaranteed to be monotic (won't go backwards). Therefore i would like to hear from you about the size of problem you were successful in running for your numerical experiments in lecture today. I would also like to revisit the Sun Platform handout that was not viewable in our "smart classroom" on Monday. We will also revisit the stommel model problem that we first examined in Lab 4 preparing to use compiler flags http://www.stewart.cs.sdsu.edu/cs575/labs/l4_flpt_compiler.html This will introduce the Jacobi iteration for solving "grid-problems" and we will contrast this with the Gauss-Seidel iteration and "discover" is parallel limitations. See you in class (and hope you plan to join the Student ACM Computer Club meeting today 4:30 pm) Kris Stewart BAM 223 - Office Hours MW 1-2:30pm