Error of pmemd.cuda ONLY within the westpa framework

[paola.b...@gmail.com]

Hi all,

I have a former westpa experience with the gromacs MD engine and the martini force field for coarse grained, but now I switched to pmemd.cuda (amber) since it runs faster on GPUs and I need a full atomic description of my system.
Although I have experience with the amber16 package, I never used into the westpa framework. I was able to initialize my westpa simulation, but I the run crashed due to a PMEMD error which occurred only within the westpa framework.

The error is the following :

Unit    9 Error on OPEN: parent.rst                                                                                                                                        

which appears tells that PMEMD is unable to open the coordinate file. However, if I run the simulation chunk from the current segment directory, the simulation runs properly.

This seems really odd and I do not understand how could pmemd be unable to open the coordinates file in the westpa framework, if it can outside the westpa framework.

Does anybody share similar experience? Any suggestion?

Thanks,

Paola

[Alex DeGrave]

Hi Paola,

I have previously run WESTPA simulations with pmemd.cuda, and I did not experience this error.  As you say, it also seems strange to me that pmemd.cuda would be unable to open files from within a WESTPA run, since the executable propagator is basically just launching your scripts (typically shell scripts).

It seems most likely to me that when WESTPA attempts to run the simulation chunk, something with your environment or shell variables is not being set the way you desire.  For example, the current working directory or your LD_LIBRARY_PATH may not be set correctly.

Would you be able to do the following to help troubleshoot?

Edit your runseg.sh (or whatever script runs pmemd.cuda) and directly before launching pmemd.cuda, add these lines:

which pmemd.cuda 

env
ls parent.rst

After making sure that STDOUT and STDERR are being logged to a file (e.g., see lines 57-58 of this west.cfg), try running the simulation again.  Based on the log files, does ls return an error? If not, do LD_LIBRARY_PATH and AMBERHOME point to the correct Amber installation directory?

Best,

Alex

--
You received this message because you are subscribed to the Google Groups "westpa-users" group.
To unsubscribe from this group and stop receiving emails from it, send an email to westpa-users+unsubscribe@googlegroups.com.
To post to this group, send email to westpa...@googlegroups.com.
For more options, visit https://groups.google.com/d/optout.

[paola.b...@gmail.com]

Hi Alex,

the 'ls' says it can't access the parent.rst file, although the file is present in the directory.
LD_LIBRARY_PATH and AMBERHOME point to the correct amber installation directory.

How can I fix it?

Thanks for your help,

Paola

To unsubscribe from this group and stop receiving emails from it, send an email to westpa-users...@googlegroups.com.

[Alex DeGrave]

Hi Paola, 

I suspect that your runseg.sh does not (or is unable to) change to the correct directory. In your log file, the "env" command likely output a line that begins "PWD=", which specifies the current working directory.  Does this point to the directory that you think it should (the directory containing the parent.rst file, and where you wish to run the segment)?  If not, you will need to add an appropriate command to change to that directory.  For example:

mkdir -pv ${WEST_SIM_ROOT}/${WEST_CURRENT_SEG_DATA_REF} || exit 1

cd ${WEST_SIM_ROOT}/${WEST_CURRENT_SEG_DATA_REF} || exit 1

(note that especially on clusters, it is advisable to run i/o intensive operations in scratch space rather than directly in $WEST_SIM_ROOT as I imply above).

If you already have such a command, do your log files show any errors from "cd"? This might be the case if, for example, you are running on a computing cluster and your desired storage array is not available from some compute nodes.

Best,

Alex

To unsubscribe from this group and stop receiving emails from it, send an email to westpa-users+unsubscribe@googlegroups.com.

[paola.b...@gmail.com]

Hi Alex,

the PWD specifies the current directory and the cd does not give me any error, but pmemd.cuda cannot open the parent.rst file.

How can't it find the file, if it is running in the proper directory?

Paola

[Ali Sinan Saglam]

Hi Paola, 

Would you mind providing a log file from your seg_logs directory after a simulation fails? Also, does the propagation fail for every segment in an iteration or only for some of them? If your runseg.sh is fairly standard it should have the environment printed and if we can see your environment variables it would be a lot easier for us. 

Best, 

Ali Sinan Saglam, PhD

Faeder lab, Room 3070, BST3

University of Pittsburgh

Pittsburgh, PA 15260

[paola.b...@gmail.com]

I attached the log file. For the purpose of debugging I am just using one replica. Shall I try more?

[Ali Sinan Saglam]

Thank you very much, this is helpful to see. I agree that everything looks correctly in place, the only thing I can currently think of doing is to manually check that the symbolic link to the bstates is correctly pointing to your eq.rst file. You can try checking if the original file you are linking to exists with something along the lines of if [ -e $PATH_TO_ORIGINAL_FILE ];then echo "original file exists";fi before linking with ln -sv in your runseg.sh.

A couple other questions, are you trying this on a cluster? If yes, are you testing it from the login node or are you submitting it as a job? 

[Paola Bisignano]

Hi all,

Thank you for help debugging. There was a typo in the path of the initial coordinate file.

Best,

Paola

To unsubscribe from this group and stop receiving emails from it, send an email to westpa-users+unsubscribe@googlegroups.com.

To post to this group, send email to westpa...@googlegroups.com.
For more options, visit https://groups.google.com/d/optout.

--

Ali Sinan Saglam, PhD

Faeder lab, Room 3070, BST3

University of Pittsburgh

Pittsburgh, PA 15260

--

You received this message because you are subscribed to a topic in the Google Groups "westpa-users" group.
To unsubscribe from this topic, visit https://groups.google.com/d/topic/westpa-users/3YAoLTno0vY/unsubscribe.
To unsubscribe from this group and all its topics, send an email to westpa-users+unsubscribe@googlegroups.com.

To post to this group, send email to westpa...@googlegroups.com.
For more options, visit https://groups.google.com/d/optout.

--

Paola Bisignano, Ph.D.
postdoctoral research associate

Department of Pharmaceutical Chemistry

Cardiovascular Research Institute
University of California, San Francisco