Page 1 of 2 1 2 >
Topic Options
#1154 - 03/31/04 07:24 PM PME in Parallel c30b1 vs Seq c27b4
jimpsull Offline
Forum Member

Registered: 02/03/04
Posts: 13
I recently switched from running c27b4 to runnin c30b1 in parallel. I was just trying to benchmark speed vs. procs on our linux cluster, beginning with two procs. I ran a script that runs successfully using c27b4 sequentially and got the following error when beginning an sd minimization with PME. Any thoughts would be appreciated

Jim Sullivan

***** LEVEL -5 WARNING FROM *****
***** FILL_CH_GRID igood .gt.LATM
******************************************
_________________________
Molecular Engineering Laboratory University of Washington

Top
#1155 - 04/06/04 06:10 PM Re: PME in Parallel c30b1 vs Seq c27b4 [Re: jimpsull]
crowley Offline
Forum Member

Registered: 11/12/03
Posts: 200
Loc: tsri (scripps) san diego, ca
Jim,
THis message says that the estimated room needed for the pme reciprocal sum was not enough. Charmm made the estimate based on the assumption that your system has a relatively homogeneous charge density. The parallel work division is in slabs of space and the arrays are dimensioned with some slack but are sized using the assumption that each slab will have about the same number of atoms.

The only way that I know of for this error message to appear is if there are too many atoms in at least one slab, meaning there is some portion of space with very low density. This happens when incorrect dimensions are used or there is a vacuum bubble in the system.

If it is this, you should be able to find the uneven density visually and decide what to do about it.
Does the error message appear with -np 1? It should not. If it does, I would
like to look more closely at your system to see if there is an error in charmm.


Hope that helps. Sincerely,
Mike
_________________________
Physical mail: Dr. Michael F. Crowley National Renewable Energy Laboratory, MS 3323 1617 Cole Blvd. Golden, CO 80401

Top
#1156 - 04/21/04 12:33 PM Re: PME in Parallel c30b1 vs Seq c27b4 [Re: crowley]
jimpsull Offline
Forum Member

Registered: 02/03/04
Posts: 13
Thank you. I do in fact have a large stretch of vacuum in my system. I used this to achieve a psuedo-2D boundary condition. For now I am proceeding without long-range electrostatics. I am justifying this on the basis of the high ionic strength (~1 molar) of my solvent. My assumption is that this ionic strength provides adequate shielding for electrostatic interactions beyond ~14 Angstroms.

Jim Sullivan
_________________________
Molecular Engineering Laboratory University of Washington

Top
#1157 - 06/16/04 03:01 PM Re: PME in Parallel c30b1 vs Seq c27b4 [Re: crowley]
rmv Offline

Forum Member

Registered: 09/17/03
Posts: 8379
Loc: 39 03 48 N, 77 06 54 W
We're doing some lipid monolayer simulations with a fair amount of vacuum space, and have encountered this problem. Is there a way in any existing CHARMM options or keywords to change the sizing of these arrays or change the slab orientation, etc.?


_________________________
Rick Venable
computational chemist


Top
#1158 - 02/22/06 05:35 PM Re: PME in Parallel c30b1 vs Seq c27b4 [Re: rmv]
rmv Offline

Forum Member

Registered: 09/17/03
Posts: 8379
Loc: 39 03 48 N, 77 06 54 W
The fix for this bug, which is still in c32 beta versions, is the following change to nbonds/pme.src at about line 410--

Before:

##IF PARALLEL
latm=max( (nattot*(3+nxyslab(0)))/nfft3*2,1)
##ELSE
latm=nattot
##ENDIF

After:


##IF PARALLEL
C latm=max( (nattot*(3+nxyslab(0)))/nfft3*2,1)
C OMIT REDUCTION FOR PARALLEL; VACUUM SPACE FIX
latm=nattot
##ELSE
latm=nattot
##ENDIF

The reduction for parallel was added for a specific machine type, which most people don't use; it isn't needed for most modern clusters.
_________________________
Rick Venable
computational chemist


Top
#23406 - 02/11/10 02:58 PM Re: PME in Parallel c30b1 vs Seq c27b4 [Re: rmv]
mooctopus Offline
Forum Member

Registered: 04/28/09
Posts: 141
Loc: U.S.A.
I have encountered the same problem. I used NPT in equilibration run, then used NVT in production run, in which I used the average of the plateau volume achieved by equilibration and also setting pgamma and internal piston pressure as 0 in all directions as recommended in Pressure.doc.

May I ask if there is anything wrong?

Thank you!

Top
#23409 - 02/11/10 03:20 PM Re: PME in Parallel c30b1 vs Seq c27b4 [Re: mooctopus]
rmv Offline

Forum Member

Registered: 09/17/03
Posts: 8379
Loc: 39 03 48 N, 77 06 54 W
What problem have you encountered? The level -5 error from the FILL_CH_GRID routine? A quick source code fix for that is shown above. Why do you ask if anything is wrong? What is your question or problem?
_________________________
Rick Venable
computational chemist


Top
#23415 - 02/11/10 04:34 PM Re: PME in Parallel c30b1 vs Seq c27b4 [Re: rmv]
mooctopus Offline
Forum Member

Registered: 04/28/09
Posts: 141
Loc: U.S.A.
My problem is after MD simulation, the waterbox was completely distorted and the volume increased enormously, and also I found the pmxx, pmyy, pmzz, pmxy,pmyz,pmxz were not 0 during dynamics. So my settings for pmass terms actually did not work.


Edited by mooctopus (02/11/10 04:34 PM)

Top
#23416 - 02/11/10 06:04 PM Re: PME in Parallel c30b1 vs Seq c27b4 [Re: mooctopus]
rmv Offline

Forum Member

Registered: 09/17/03
Posts: 8379
Loc: 39 03 48 N, 77 06 54 W
That has nothing to do with this bug report post. Post something more relevant and detailed (input script, significant output messages) in the appropriate forum. I do NVT using CPT with PCONS PINT PMASS 0. all of the time; there is no bug. You can also simply use CPT with HOOVER (w/o PCONS PINT PMASS 0.) to do NVT, but you do not get reporting of P data that way.
_________________________
Rick Venable
computational chemist


Top
#23417 - 02/12/10 12:56 PM Re: PME in Parallel c30b1 vs Seq c27b4 [Re: rmv]
mooctopus Offline
Forum Member

Registered: 04/28/09
Posts: 141
Loc: U.S.A.
This is my script. Can I use both Hoover and Pcons Pint pmass 0?
I thought the Hoover is for temperature control.
btw, is that the same by using "cpt" and "cpt leap"?

=====scritp===
Dynamics rest cpt timestep 0.001 nstep 1000000 -
nprint 100 iprfrq 500 -
PCONst pmass 0.0 pgamma 0.0 PREFerence 1.0 -
HOOVer tmass 1000 reft 323 -
EWALD PMEWALD KAPPA 0.32 FFTX 90 FFTY 90 FFTZ 90 ORDER 6 -
CUTNB 14.0 CTONNB 10.0 CTOFNB 12.0 VDW VSWITCH NTRFRQ 10 -
inbfrq -1 ihbfrq 0 imgfrq -1 -


Edited by mooctopus (02/12/10 12:59 PM)

Top
Page 1 of 2 1 2 >

Moderator:  BRBrooks, bucknerj, chmgr, John Legato