RE: [gmx-users] Re: trouble parallelizing a simulation over a cluster
Thanks Carsten for the hint, you were exactly right, the mdrun_mpi was not linked to the openmpi library. I corrected this by recompiling the mdrun_mpi after issuing the following command: module load openmpi/gnu Now it is running very well Hassan From: gmx-users-boun...@gromacs.org on behalf of Carsten Kutzner Sent: Wed 12/8/2010 4:16 AM To: Discussion list for GROMACS users Subject: [gmx-users] Re: trouble parallelizing a simulation over a cluster On Dec 8, 2010, at 1:03 PM, Hassan Shallal wrote: > Thanks a lot Justin for the very helpful answers concerning the pressure > equilibration. Using Berendsen Barostat over 200 ps has lead to the correct > average pressure... > > I have another issue to discuss with you and with the Gromacs mailing list > members; > > I have been trying to run a simulation on a computer cluster for the first > time using a sub file script. What happened is that the .sub file attempted > to run the simulation 24 times instead of parallelizing it over the 24 > processors > > Here are the contents of run_1.sub file I tried to use to parallelize the > simulation using qsub run_1.sub > > #PBS -S /bin/bash > #PBS -N run_1 > #PBS -l nodes=3:ppn=8 > module load openmpi/gnu > mpirun -np 24 /home/hassan/bin/bin/mdrun_mpi -deffnm run_1 -v &> > run_1_update.txt > exit $? > What happens it that it outputs 24 run_1.log files, starting from > #run_1.log1# all the way to #run_1.log23#...Has anyone faced this problem > before? and If yes, any hints or solutions? This typically happens when using a serial mdrun. You should check with ldd whether mdrun_mpi is linked to the correct mpi library. Carsten > > I do appreciate any help in that domain > Hassan > > From: gmx-users-boun...@gromacs.org on behalf of Justin A. Lemkul > Sent: Mon 12/6/2010 6:43 PM > To: Discussion list for GROMACS users > Subject: Re: [gmx-users] pressure fluctuations > > > > Hassan Shallal wrote: > > Dear Gromacs users, > > > > I have some concerns about the both the pressure fluctuations and > > averages I obtained during the equilibration phase. I have already read > > through several similar posts as well as the following link > > http://www.gromacs.org/Documentation/Terminology/Pressure. I understand > > the pressure is a macroscopic rather than instantaneous property and the > > average is what really matters. I also found out through similar posts > > that negative average pressure indicates the system tendency to contract. > > > > In the above link, it mentioned that pressure fluctuations should > > decrease significantly with increasing the system's size. In my cases, I > > have a fairly big systems (case_1 with *17393* water molecules > > and case_2 with *11946 *water molecules). However, the pressure still > > has huge fluctuations (around 500 bars) from the reference value (1 > > bar). Here are the average pressure and density values resulting from > > the equilibration phases of two cases, please notice the negative > > average pressure values in both cases... > > > > Case_1_pressure: > > Energy Average Err.Est. RMSD Tot-Drift > > --- > > Pressure *-2.48342* 0.92369.709 -4.89668 > > (bar) > > Case_1_density: > > Energy Average Err.Est. RMSD Tot-Drift > > --- > > Density 1022.89 0.38 3.82532.36724 > > (kg/m^3) > > Case_2_pressure: > > Energy Average Err.Est. RMSD Tot-Drift > > --- > > Pressure *-8.25259*2.6423.681 -12.1722 > > (bar) > > Case_2_density: > > Energy Average Err.Est. RMSD Tot-Drift > > --- > > Density 1034.11 0.372.499641.35551 > > (kg/m^3) > > > > So I have some questions to address my concerns: > > 1- each of the above systems has a protein molecule, NaCl to give 0.15 M > > system and solvent (water) molecules... Could that tendency to contract > > be an artifact of buffering the system with sodium and chloride ions? > > > > I suppose anything is possible, but given that these are fairly standard > conditions for most simulations, I tend to doubt it. My own (similar) systems > do not sho
[gmx-users] Re: trouble parallelizing a simulation over a cluster
On Dec 8, 2010, at 1:03 PM, Hassan Shallal wrote: > Thanks a lot Justin for the very helpful answers concerning the pressure > equilibration. Using Berendsen Barostat over 200 ps has lead to the correct > average pressure... > > I have another issue to discuss with you and with the Gromacs mailing list > members; > > I have been trying to run a simulation on a computer cluster for the first > time using a sub file script. What happened is that the .sub file attempted > to run the simulation 24 times instead of parallelizing it over the 24 > processors > > Here are the contents of run_1.sub file I tried to use to parallelize the > simulation using qsub run_1.sub > > #PBS -S /bin/bash > #PBS -N run_1 > #PBS -l nodes=3:ppn=8 > module load openmpi/gnu > mpirun -np 24 /home/hassan/bin/bin/mdrun_mpi -deffnm run_1 -v &> > run_1_update.txt > exit $? > What happens it that it outputs 24 run_1.log files, starting from > #run_1.log1# all the way to #run_1.log23#...Has anyone faced this problem > before? and If yes, any hints or solutions? This typically happens when using a serial mdrun. You should check with ldd whether mdrun_mpi is linked to the correct mpi library. Carsten > > I do appreciate any help in that domain > Hassan > > From: gmx-users-boun...@gromacs.org on behalf of Justin A. Lemkul > Sent: Mon 12/6/2010 6:43 PM > To: Discussion list for GROMACS users > Subject: Re: [gmx-users] pressure fluctuations > > > > Hassan Shallal wrote: > > Dear Gromacs users, > > > > I have some concerns about the both the pressure fluctuations and > > averages I obtained during the equilibration phase. I have already read > > through several similar posts as well as the following link > > http://www.gromacs.org/Documentation/Terminology/Pressure. I understand > > the pressure is a macroscopic rather than instantaneous property and the > > average is what really matters. I also found out through similar posts > > that negative average pressure indicates the system tendency to contract. > > > > In the above link, it mentioned that pressure fluctuations should > > decrease significantly with increasing the system's size. In my cases, I > > have a fairly big systems (case_1 with *17393* water molecules > > and case_2 with *11946 *water molecules). However, the pressure still > > has huge fluctuations (around 500 bars) from the reference value (1 > > bar). Here are the average pressure and density values resulting from > > the equilibration phases of two cases, please notice the negative > > average pressure values in both cases... > > > > Case_1_pressure: > > Energy Average Err.Est. RMSD Tot-Drift > > --- > > Pressure *-2.48342* 0.92369.709 -4.89668 > > (bar) > > Case_1_density: > > Energy Average Err.Est. RMSD Tot-Drift > > --- > > Density 1022.89 0.38 3.82532.36724 > > (kg/m^3) > > Case_2_pressure: > > Energy Average Err.Est. RMSD Tot-Drift > > --- > > Pressure *-8.25259*2.6423.681 -12.1722 > > (bar) > > Case_2_density: > > Energy Average Err.Est. RMSD Tot-Drift > > --- > > Density 1034.11 0.372.499641.35551 > > (kg/m^3) > > > > So I have some questions to address my concerns: > > 1- each of the above systems has a protein molecule, NaCl to give 0.15 M > > system and solvent (water) molecules... Could that tendency to contract > > be an artifact of buffering the system with sodium and chloride ions? > > > > I suppose anything is possible, but given that these are fairly standard > conditions for most simulations, I tend to doubt it. My own (similar) systems > do not show this problem. > > > 2- how to deal with the tendency of my system to contract? Should > > I change the number of water molecules in the system? > > or > > Is it possible to improve the average pressure of the above systems by > > increasing the time of equilibration from 100 ps to may be 500 ps or > > even 1 ns? > > > > 3- Is there a widely used range of average pressure (for ref_p = 1 bar) > > that indicates acceptable equilibration of the system prior to the > > production? > > > > To answer #2 and #3 simultaneously - equilibration is considered "finished" > when > your system stabilizes at the appropriate conditions (usually temperature and > pressure). Your results indicate that your equilibrium is insufficient. > > > 4- I can't understand how the system has a tendency to contract whereas > > the average density of the solvent is already slightly higher than it
