[gmx-users] Re: reasons for slow computation?

[Mark Abraham]

On 12/10/2011 3:48 AM, Yun Shi wrote:
> Hi Mark,
>
> I am not quite sure. While the "details" are different, I guess it is 
> still valid to compare some statistical properties, right? For 
> example, I should still be able to compare the average COM distance of 
> these two ligands over long trajectories, and make the conclusion that 
> one is better than the other since the distance is smaller. So should 
> I be able to compare any averaged, or say, clustered properties?

It's like comparing a bench experiment done with the same method on 
Monday by Bill with one done on Tuesday by Mary. Each of them stirred 
their reaction vessel a different number of times under a different 
phase of the moon, but if full mixing of the reagents occurred both 
times, the results can be compared.

Mark

>
> Thanks,
> Yun
>
>
> On 11/10/2011 1:40 PM, Yun Shi wrote:
> > Hi Justin,
> >
> > I guess you are right, that some processors on that cluster appear to
> > be much slower than others.
>
> More likely is that the mapping of processes to processors is faulty, as
> Justin said.
>
> >
> > But I am still wondering that, would the difference in initial maximum
> > inter charge-group distances (0.451 nm vs 0.450 nm) and minimum
> > initial size of DD gird (0.620nm vs 0.618nm) make the two simulations
> > NOT comparable? Because I want to compare the interactions of two
> > different ligands with the same protein.
>
> That's just a detail of how the simulation is being implemented in
> parallel. There are a vast number of algorithmically equivalent ways to
> do this, and the initial conditions determine which is chosen. Since
> your initial conditions differ, so do the details.
>
> Mark
>
>

-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://maillist.sys.kth.se/pipermail/gromacs.org_gmx-users/attachments/20111012/52a24118/attachment.html>