[gmx-users] A few general questions on Gromacs

[Martin Höfling]

Am Montag, 25. Juni 2007 schrieb Corina Stratan:

Hi Cornia,

these are my experiences, so others might differ from mine:

> 1) Which is the typical size of the input/output files for Gromacs
> programs? The ones included in the tutorial are below 50K; are the files
> larger in "real" situations? And if yes, is there any place we can find
> some larger test files?

If you consider the PDB or binary topology (tpr) file as input, you are in the 
order of 100k-1M. This size mainly depends on the size of your system. 
Parameters contribute pretty much nothing to the input size.
Same with the output: it depends what you're interested on. For some 
applications, a single (uncompressed) trajectory grew to 100G, but typically 
it's at least one magnitude below. If you want a _large_ input, you can just 
solvate a protein in a large water box. 

> 2) Which would be the typical running time of an mdrun simulation? (or put
> it another way, is the runtime usually significant - i.e., more than a few
> minutes?)

This depends on what you're interested in. For parameter studies, where you 
need hundreds or thousands of simulations, you usually use small systems. 
Runtime is in the order of minutes to hours, sometimes in days. (on a single 
processor)
For other experiments, simulation time can grow to 3 month for one simulation 
on a single processor)

> 3) Are the jobs usually run in batch? And if yes, do the batches contain a
> single program, run with different input files, or multiple programs?

Normally, you run "mdrun" together with different input files. What do you 
mean by multiple programs? Batch scripts sometimes contain preprocessing 
(with different programs) but this is executed in serial and my guess would 
be that mdrun takes 99.99% of cpu time for most applications. If you consider 
MPI processes on different machines as individual programs, then there can be 
a lot, depending on parallelization level used.

Best 
	Martin