[gmx-users] About the binary identical continuation by restarting from the checkpoint file

[Berk Hess]

Hi,

Did you redo the run you continued from as well with 4.0.4?

Also there are two other issues that can prevent binary identical contiuation:
dynamic load balancing and FFT optimization (when using PME with fftw).

Dynamic load balancing can be turned off with mdrun -dlb no
You can turn off everything that affect reproducability, including fftw,
with mdrun -reprod

I compile fftw with SSE and then do for Gromacs:
configure --disable-fftw-measure
to ensure reproducible simulations with PME.

Berk

From: guogj at hotmail.com
To: gmx-users at gromacs.org
Subject: RE: [gmx-users] About the binary identical continuation by	restarting	from the checkpoint file
Date: Wed, 25 Feb 2009 02:43:38 +0000








Hi,

 

I change to Gromacs 4.0.4 now, and perform NPT with four CPU cores. The results of continuation runs are still different from the undivided run. Neither phenomenon described previously disappears.

 

Guang-Jun

 



From: gmx3 at hotmail.com
To: gmx-users at gromacs.org
Subject: RE: [gmx-users] About the binary identical continuation by restarting from the checkpoint file
Date: Tue, 24 Feb 2009 14:44:07 +0100



Hi,

Are you running NPT, single processor with 4.0.3?
Then you are experiencing the bug that the box is not updated in the checkpoint file.
Please change to 4.0.4.

Berk



From: guogj at hotmail.com
To: gmx-users at gromacs.org
Subject: RE: [gmx-users] About the binary identical continuation by restarting from the checkpoint file
Date: Tue, 24 Feb 2009 08:37:38 +0000



Hi, Mark,
 


I reduce the length of test runs to only 6 steps now. The binary identical continuation still cannot be obtained. I use gmxcheck to compare the results according to your suggestions and I find two reasons to cause different trajectories. 
 
One is a water molecule happening to span the Y-boundary of the simulation box. In the undivided run, the molecule locates at the +Y direction while it occurs at the ¨CY direction in the continuous run. The differences of Y-coordinates are exactly the box length. Certainly, X and Z coordinates in the two runs are indeed identical. 
 
The other one is the velocities of many atoms being different in the fifth or sixth digits. The case becomes more and more serious with time. 
 
After I change the different initial gro files, the first phenomenon disappears because no spanning-boundary water occurs. However, the second one is always there no matter which Gromacs is used, the single-precision or the double-precision. So, my question is how to obtain the binary identical continuation. 
 
regards,
Guang-Jun 
 

> Date: Mon, 23 Feb 2009 14:37:06 +1100
> From: Mark.Abraham at anu.edu.au
> To: gmx-users at gromacs.org
> Subject: Re: [gmx-users] About the binary identical continuation by restarting from the checkpoint file
> 
> GuoGuangjun wrote:
> > Hi, All
> > 
> > 
> > 
> > Before to perform long simulations, I think it is necessary to do 
> > continuous runs. I make several tests
> > 
> > to check the reliability of restart by using the checkpoint file. First, 
> > I do a 6ps-long run, and then do it
> > 
> > again by two parts, that is, the first half (3 ps) starts from the same 
> > initial gro file, followed by its
> > 
> > second half. Here are my operations:
> > 
> > 
> > 
> > grompp -f 6ps1.mdp -c water.gro -p water.top -o 6ps1.tpr
> > 
> > mpirun -np 4 mdrun -s 6ps1.tpr -o 6ps1.trr -x 6ps1.xtc -e 6ps1.edr -dlb no
> > 
> > 
> > 
> > grompp -f 3ps.mdp -c water.gro -p water.top -o 3ps.tpr
> > 
> > mpirun -np 4 mdrun -s 3ps.tpr -o 6ps2.trr -x 6ps2.xtc -e 6ps2.edr -dlb 
> > no -cpo 3ps.cpt
> > 
> > 
> > 
> > tpbconv -s 3ps.tpr -extend 3.0 -o 6ps2.tpr -cont
> > 
> > mpirun -np 4 mdrun -s 6ps2.tpr -o 6ps2.trr -x 6ps2.xtc -e 6ps2.edr -dlb 
> > no -cpi 3ps.cpt -append
> > 
> > 
> > 
> > The log file tells me----¡°Restarting from checkpoint, appending to 
> > previous log file.¡± However, when I
> > 
> > check the binary files, such as 6ps1.trr and 6ps2.trr, the cmp command 
> > tells me they are not identical.
> 
> Use gmxcheck for this test. This will give you either confirmation that
> you're achieving your objective, or enough information to find out why not.
> 
> > During these tests, I have turned off the gen_vel, optimize_fft, and 
> > dynamic load balancing (dlb) according
> > 
> > to Gromacs manual and previous posters. How to obtain the binary 
> > identical continuation? My computational
> > 
> > environments are Gromacs 4.0.3, lammpi 7.1.4, Redhat Linux, two AMD 
> > Opteron CPUs with dual cores
> > 
> > in each, 500 SPC water molecules, and the NPT ensemble.
> 
> Your approach looks right to me.
> 
> Mark
> _______________________________________________
> gmx-users mailing list gmx-users at gromacs.org
> http://www.gromacs.org/mailman/listinfo/gmx-users
> Please search the archive at http://www.gromacs.org/search before posting!
> Please don't post (un)subscribe requests to the list. Use the 
> www interface or send it to gmx-users-request at gromacs.org.
> Can't post? Read http://www.gromacs.org/mailing_lists/users.php



¸ü¶àÈÈÀ±×ÊѶ¾¡ÔÚаæMSNÊ×Ò³£¡ Á¢¿Ì·ÃÎÊ£¡


Express yourself instantly with MSN Messenger! MSN Messenger
ʹÓÃÐÂÒ»´ú Windows Live Messenger ÇáËɽ»Á÷ºÍ¹²Ïí£¡ Á¢¿ÌÏÂÔØ£¡
_________________________________________________________________
Express yourself instantly with MSN Messenger! Download today it's FREE!
http://messenger.msn.click-url.com/go/onm00200471ave/direct/01/
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://maillist.sys.kth.se/pipermail/gromacs.org_gmx-users/attachments/20090225/7b25eb7e/attachment.html>