From: Peter Freddolino (petefred_at_ks.uiuc.edu)
Date: Tue Sep 05 2006 - 12:14:49 CDT
Hi Brian,
you might want to try this multiple times, and on a larger/longer job;
my guess is on a job this length the startup cost is a significant
percentage of your runtime, and this is going to be quite variable and
may be slower in newer versions even if the run itself is faster (this
is also likely to be very variable depending on when you're running).
Peter
On Tue, Sep 05, 2006 at 10:02:29AM -0700, Brian Bennion wrote:
>
> Hello,
>
> I have been comparing timing differences between NAMD 2.5 and 2.6 on the
> apoa1 benchmark.
> I noticed something strange. The secs/step for 32 cpus decrease
> 0.077 to 0.070 which I thought was great, however, the final wall clock
> actually increase from 51.43s (namd2.5) to 55.11s (namd2.6) for the same
> 32 cpus.
>
> Then I looked at the last timing output statement at step 500 and saw that
> namd2.5 was 45.03s while namd2.6 was 41.72s. So the ouputs during runtime
> seem to be congruent, but the final wall clocks don't follow in the same
> direction.
>
>
> My questions are then, which metric should be trusted and secondly why are
> the final wall clock values backwards?
>
> Thank you for your help.
>
> brian
>
> ************************************************
> Brian Bennion, Ph.D.
> Biosciences Directorate
> Lawrence Livermore National Laboratory
> P.O. Box 808, L-446 bennion1_at_llnl.gov
> 7000 East Avenue phone: (925) 422-5722
> Livermore, CA 94550 fax: (925) 424-5513
> ************************************************
This archive was generated by hypermail 2.1.6 : Wed Feb 29 2012 - 15:43:58 CST