GROMACS version: 2019.2
GROMACS modification: No
I’m running long simulations using GROMACS on local HPC. I’m suspecting that the performance of the run is changing over time probably due to some external processes or configurations. What’s worse when I run out of the allocated time application is hard killed so usually there is no performance info at the end of the log.
Is there any way to force GROMACS to more frequently log performance statistics during runs?
Thanks!
If you run in verbose mode (mdrun -v) you will have performance (if you run with nsteps = -1) or remaining time printed to the standard output. If this output is redirected to a file, the former will help out see whether there is any performance variation.
I know about the “-v” option but it only prints statistics at the end of the run.
I had in mind something more in NAMD style so printing benchmark line every n steps
i.e:
Info: Benchmark time: 47 CPUs 0.0475851 s/step 0.275377 days/ns 13540 kB memory
TIMING: 1000 CPU: 18.35, 0.01831/step Wall: 50.1581, 0.0499508/step, 6.92374 hours remaining, 14244 kB of memory in use.
The way I figured it out is that every time checkpoint is created gromacs logs steps and timestamp base on that I can monitor current ns/day performance
The performance is reported every -stepout number of steps (default 100), not only at the end. Thus, you should get very frequent (or at your option, less frequent) updates.