On Thu, Aug 25, 2011 at 6:02 PM, Chris Bryant <csb61.case.edu> wrote:
> Hi,
>
> I am trying to estimate how many cpu hours a simulation will take, but I am
> not sure how to calculate it. The output file gives time began and ended,
> which I could use to get the total simulation time and then multiply that
> by
> cpus to get an upper bound on cpu hours (assuming the total simulation time
> was not at 100% efficiency), but I was wondering if their was a better
>
That should be your exact CPU hours (number of processors * wallclock time)
as charged by a supercomputing center. If you reserve 12 processors to run
a serial calculation on (so 11 are running idle), you're still charged or
those 12 processors, even though 11 are running near 0% capacity. I think
that's the best you can do.
> indication of this. There is another time given in the output files,
> labeled "total time," but I am not sure what units it is in. For instance,
> a simulation run with 16 cpus with a "total time" of 37737 clocks in at
>
37737 is in seconds. If you divide it by 3600, you get 10.4825, or
10:28:57. Run on 16 CPUs, this is 167.72 CPU-hours. That's if you don't
include any other time taken (i.e. if there's anything else in your script
that might take some time).
HTH,
Jason
--
Jason M. Swails
Quantum Theory Project,
University of Florida
Ph.D. Candidate
352-392-4032
_______________________________________________
AMBER mailing list
AMBER.ambermd.org
http://lists.ambermd.org/mailman/listinfo/amber
Received on Thu Aug 25 2011 - 16:30:04 PDT
