Re: Network version gives inconsistent wall clock timings

From: Jim Phillips (jim_at_ks.uiuc.edu)
Date: Wed Apr 13 2011 - 09:04:46 CDT

How are you measuring wallclock time? If you look at the periodically
printed TIMING output lines is the time/step steady or does the ibverbs
build lock up at some point?

-Jim

On Wed, 30 Mar 2011, yashodhan karandikar wrote:

> I am running NAMD network version compiled with ibverbs for infiniband, with
> Lysozyme input for 250000 steps. It gives highly inconsistent WallClock
> times across multiple runs (for example one run took 2500 seconds, another
> run with the same input took 6600 seconds). However, when i ran NAMD mpi
> version with Hp-mpi for the same input, it is giving consistent WallClock
> timings across multiple runs. Also, the days/ns reported by both versions
> remain stable across multiple runs.
>
> Does anybody have any idea as to why this might be happening ?
>
> Thanks.
>
> Regards,
> Yashodhan Karandikar
>

This archive was generated by hypermail 2.1.6 : Mon Dec 31 2012 - 23:20:07 CST