[gmx-users] A question on execusion speed

[Mark Abraham]

----- Original Message -----
From: "Justin A. Lemkul" <jalemkul at vt.edu>
Date: Wednesday, May 26, 2010 3:16
Subject: Re: [gmx-users] A question on execusion speed
To: Discussion list for GROMACS users <gmx-users at gromacs.org>

> 
> 
> Paymon Pirzadeh wrote:
> >Hello,
> >I have question regarding the execution time of a trajectory. I 
> used to
> >run my trajectory at 265K or 275K and at the end of the log 
> file the
> >speed was reported as e.g. 1.18 ns/day. Recently, I changed the
> >temperature to 288K and now my simulation is not finished in the
> >assigned wall time (240 hours). The potential and total energy 
> and other
> >parameters in the log file looks fine. Is there really sth 
> going on with
> >temperature or the cluster I am running the jobs on?!
> 
> I can't think of a reason why simply changing the temperature 
> would drastically alter the speed of the simulation.  Look 
> to the hardware.  When I've had random decreases in 
> performance, it's because a node is having issues or the 
> interconnect is bugging out.

Indeed. Obviously one should use gmxcheck to compare the two .tpr, and be confident the options to mdrun and the job script conditions were constant, before casting nasturtiums!

Mark