Measuring OpenMP performance

Use this forum to discuss the book: Using OpenMP - Portable Shared Memory Parallel Programming, by Barbara Chapman, Gabriele Jost and Ruud van der Pas Read the viewtopic.php?f=8&t=465 for book info and to download the examples. Post your feedback about the book and examples to this forum

Measuring OpenMP performance

Postby Stephane » Sat Jan 02, 2010 8:19 am

Hello, I have a few questions about the book:

on page 139, formula 5.1, is Tcpu stand for the TOTAL cpu time? I assume it is, even if it is not clearly pointed out.

Also, at the end of the same page, why are you saying that "the performance on four processors is already far from optimal"? I mean, it's the elapsed time that really matter in the end, isn't it?

Thank you for this excellent book,

Stephane
Stephane
 
Posts: 25
Joined: Fri Dec 18, 2009 10:04 am

Re: Measuring OpenMP performance

Postby ruud » Wed Jan 06, 2010 4:01 am

Hi Stephane,

Thank you very much for the feedback. I'm very glad to hear you like the book :)

You are correct in assuming that Tcpu represents the total CPU time. As you suggest, it would have been better to clearly state this.

In your second comment you bring up an interesting point. It kind of depends on the perspective and expectation. From a parallel efficiency point of view, the performance is poor, indicating many cycles are wasted if this was the only program using the processors. At the end of the day however, the elapsed time is indeed still less compared to using 2 processors. Since the cost of cores/threads is coming down, a relatively low efficiency is not so much of a concern as long as there is a reduction in elapsed time. That is at least what I notice when talking with people about efficiency versus elapsed time.

Kind regards, Ruud
ruud
 
Posts: 23
Joined: Mon Nov 26, 2007 2:13 am


Return to Using OpenMP - The Book and Examples

Who is online

Users browsing this forum: No registered users and 0 guests