[gmx-users] Does mdrun know when other gromacs programs move its destination .xtc?

[chris.neale at utoronto.ca]

When an output .xtc destination of mdrun is backed up while the original version
of mdrun is still running, does mdrun know and does the output get handled
correctly?

here is what happened to me:

1) mdrun produces traj.xtc
2) Beleiving that mdrun had died due to a hardware error, I restarted the run
from 2.5ns by way of tpbconv
3) The output from the restart indicates: Back Off! I just backed up traj.xtc to
./#traj.xtc.2#
4) Now, 2 weeks later I realize that the initial run never actually died, it was
just hidded from ps2 because of the queue system on our cluster, which I didn't
fully understand until now.

So two versions of mdrun are running simultaneously in the same directly
(identified by the original log file which continues to show additional md
steps). The strange thing is that #traj.xtc.2# continues to be updated and is
now the size that I expect it to be as if it is getting properly updated.
Is this possible? Does mdrun know when "Back Off!" occurs and redirect its
output appropriately?

The restart was 2 weeks ago and here is an ls -hl from today:
1.8G Apr 13 11:41 #traj.xtc.2#
1.4G Apr 13 11:41 traj.xtc

A colligue suggested that perhaps both trajectories are corrupted since traj.xtc
may be continually backed up to ./#traj.xtc.2# in a ping-pong type manner as
each version of mdrun tried to write to the same file. They then suggested that
this was not likely as the numbering on the backed up file should increase.

Thanks for any help.
I am using gromacs-3.3 on intel linux machines.
Both versions of mdrun continue to run.