[gmx-users] Domain decomposition in FEP

[Mark Abraham]

On 15/02/2012 1:44 AM, lloyd riggs wrote:
> Dear Alexey Zeifman,
>
> > From my experience it is just the nodes have to break down into some sort of even multiple by dividing the domain decomposition size or vise versi.  I think somone elses answere was just to try it with a couple sizes for the decomposition, as multiples of the number of nodes.
>
> also your PME reserved nodes are probably too much or your wasting space, but I am not sure if this effects the domain decomposition or if all 20 nodes are thrown into that?

This is not really the issue. You can read in the OP's output that there 
are 20 PP and 12 PME nodes, which is a normal enough ratio, and fine for 
a 0.29 PME load...

If the OP is confident that his initial configuration really doesn't 
have such a long bond (no input checkpoint file messing things up?) he 
should try again with GROMACS 4.5.5.

Mark

>
> Stephan Watkins
>
> -------- Original-Nachricht --------
>> Datum: Tue, 14 Feb 2012 18:01:24 +0400
>> Von: Alexey Zeifman<azeif at mail.ru>
>> An: gmx-users at gromacs.org
>> Betreff: [gmx-users] Domain decomposition in FEP
>> Dear all,
>>
>> I'm trying to run FEP calculation of the ligand in the protein in water.
>> Performing NVT-dynamics for Lennard-Johnes perturbation using soft core
>> results in the following error:
>>
>> There is no domain decomposition for 20 nodes...
>>
>> I've checked out the log-file and found this:
>>
>>> Initial maximum inter charge-group distances:
>>>      two-body bonded interactions: 1.474 nm, Bond, atoms 4371 4391
>>   >  multi-body bonded interactions: 0.522 nm, Ryckaert-Bell., atoms 3747
>> 3756
>>> Minimum cell size due to bonded interactions: 1.621 nm
>>> Guess for relative PME load: 0.29
>>> Will use 20 particle-particle and 12 PME only nodes
>>> This is a guess, check the performance at the end of the log file
>>> Using 12 separate PME nodes
>>> Scaling the initial minimum size with 1/0.8 (option -dds) = 1.25
>>> Optimizing the DD grid for 20 cells with a minimum initial size of 2.026
>> nm
>>> The maximum allowed number of cells is: X 3 Y 3 Z 3
>>>
>> ->------------------------------------------------------
>>> Program mdrun_mpi, VERSION 4.5.1
>>> Source code file: domdec.c, line: 6428
>>> Fatal error:
>>> There is no domain decomposition for 20 nodes that is compatible with the
>> given box and a minimum cell size of 2.02621 nm
>>> Change the number of nodes or mdrun option -rdd or -dds
>>> Look in the log file for details on the domain decomposition
>>> For more information and tips for troubleshooting, please check the
>> GROMACS
>>> website at http://www.gromacs.org/Documentation/Errors
>> ->------------------------------------------------------>
>>
>> I understand that distance of 1.474 nm is rather big for bond! BUT I can't
>> find such a distance between atoms 4371 4391. The fragment of gro-file for
>> such index interval looks like this:
>>
>>    640LIG     C4 4371   2.897   3.416   3.072
>>    640LIG     N3 4372   3.296   3.109   3.434
>>    640LIG     N6 4373   3.416   2.999   3.595
>>    640LIG     C6 4374   2.696   2.762   3.601
>>    640LIG     C7 4375   3.025   3.232   3.150
>>    640LIG     C8 4376   3.208   3.142   3.338
>>    640LIG     C9 4377   2.758   2.644   3.649
>>    640LIG    C10 4378   2.767   2.884   3.595
>>    640LIG     N4 4379   3.319   2.826   3.731
>>    640LIG    C11 4380   2.968   2.768   3.677
>>    640LIG    C12 4381   2.894   2.648   3.686
>>    640LIG    C13 4382   2.901   2.887   3.638
>>    640LIG     N5 4383   3.100   2.766   3.707
>>    640LIG    C14 4384   3.420   2.908   3.693
>>    640LIG     O1 4385   3.018   3.627   3.068
>>    640LIG    C15 4386   3.028   3.095   3.182
>>    640LIG    C16 4387   3.123   3.049   3.275
>>    640LIG     O2 4388   2.566   2.759   3.559
>>    640LIG     O3 4389   2.699   2.993   3.545
>>    640LIG     O4 4390   2.680   2.530   3.655
>>    640LIG    C17 4391   2.791   3.426   3.184
>>    640LIG    C18 4392   2.834   3.453   2.937
>>
>> and the actual distance is 1.55 A - quite OK for single C-C bond.
>>
>> And another interesting fact is that it doesn't occur during FEP
>> calculation of the same system without soft-core and aimed at only charge
>> perturbation.
>>
>> I will appreciate any help regarding this.
>>
>> Alexey Zeifman
>>