Simulation freezing while reading Plumed script

openmm / openmm-plumed

OpenMM plugin to interface with PLUMED

55 stars 23 forks source link

Simulation freezing while reading Plumed script #51

Closed mdpoleto closed 2 years ago

mdpoleto commented 3 years ago

Hi,

I have been facing this problem for a while and I could not find a solution for it. I ran a 500ns metadynamics on my system using both OpenMM and Plumed, which was not enough to achieve convergence of my FES. I want to extend my simulation, but my simulation freezes while reading the Plumed script as shown on Production.log below.

Production.log

My plumed script is intended to create a cylinder volume based metadynamics, as below. metady_cyl_K2.txt

My OpenMM script is this: Production_script.zip

I have seen this freezing behavior in different systems and metadynamics protocols as well. I have faced this issue while running locally in my PC and in a HPC facility.

Any ideas?

raimis commented 3 years ago

Hi @mdpoleto,

Thank for reporting the issue.

What does the "freezing behavior" means exactly? Does CPU doing some work or its idle?
I don't see any error messages? Have you recorded stderr too?
Could you include all the files (including the system files) necessary to reproduce the issue?
Is it possible to reproduce the issue quickly without running a 500 ns simulation?

mdpoleto commented 3 years ago

Hi @raimis

I followed your suggestion of checking the reproducibility of the issue in shorter simulations and I figured out the problem. In shorter simulations (and smaller systems), the time Plumed takes to read is shorter, but I could detect it.

It turns out my Plumed script has the flag CALC_RCT on its METAD protocol and, while restarting my simulation, Plumed takes some time to re-read all deposited gaussians in order to continue its calculation of CALC_RCT. Since my system is somewhat big and it had ran for 500ns, it took ~20min to read everything. Afterwards, the simulation continued without any problem.

Thank you for you suggestions and I am sorry for the inconvenience.

raimis commented 2 years ago

@mdpoleto it is good to hear that it works.

Unfortunately, there is nothing we can do at the OpenMM-PLUMED level to increase the reading speed of PLUMED. If you report at https://github.com/plumed/plumed2, maybe they can improve it.