[gmx-users] errors on restart with tpbconv

[Mark Abraham]

gianluca santarossa wrote:
> Florian Haberl wrote:
> 
>> Hi,
>>
>> if the run died before wrting out a complete frame you have to restart 
>> from the last complete frame.
>>   
> 
> I agree :), but is there a way to check out automatically the last 
> useful frame?
> If, as I think, ener.edr is written after (or less frequently than) 
> traj.trr, than traj.trr has equal or more frames than ener.edr. 
> Therefore, why gromacs reads traj.trr before ener.edr? And why it 
> computes the last frame from traj.trr?
> I really don't understand how to overcome the problem. I think there 
> should be a way to directly compute the last useful frame. Or to update 
> the energy files more frequently, therefore minimizing the loss of data.

You are in control of how frequently .trr and .edr files are written (in 
simulation time) with options to the .mdp file. See the manual. You are 
not in control of how often in real time the buffered I/O is flushed to 
disk. The only code-base solution would be to flush the buffers all the 
time, which defeats the point of buffering I/O.

Your observations here are consistent with the .trr file - which gets 
written in larger chunks than the .edr file - always filling its buffer, 
but .edr only occasionally. While this kind of observation will be a 
frequent occurrence in the case of catastrophic crashes, I think the 
"convenience" of a more automated restart along the lines you suggest is 
outweighed by the times the automation will do something the user 
doesn't anticipate, and worse, never knew happened. I think it is better 
policy to give this error message and force the user to think about 
exactly what they want to do.

Mark