[molpro-user] "embarrassingly parallel" ccsd geometry optimization

[Andy May]

Jan,

The output you sent was not the Molpro output, but rather the job 
scheduler output, so there is no way to see where the failure happened.

I ran your input file, but it's quite a large calculation so it exceeded 
the maximum walltime of the cluster I was running on, but up to 72 hours 
it was running with no problem.

The error you report is quite often seen when the scratch disk is full, 
perhaps this was the case or there were some other jobs running also 
filling up the disk at the time?

Best wishes,

Andy

On 27/12/11 18:55, Jan Steckel wrote:
> Molpro-Users
>
> This question concerns an "embarrassingly parallel" ccsd geometry
> optimization (job.tar attached).
>
> serial job: first ccsd 8 hours and 8 hours each for ccsd evaluations in
> OPTG
> 8 core job: first ccsd 11 hours and 8 hours each for ccsd evaluations in
> OPTG
> 16 core job: first ccsd 53 hours and 8 hours each for ccsd evaluations
> in OPTG
>
> No matter whether the job is run on 1, 8 or 16 cores, sooner or later
> there is a read or a write error. (write error in iow_directr_write)
>
> 1) Does anyone have suggestions on how to avoid the disk read/write errors?
> 2) Is it possible for molpro to run the first ccsd evaluation in serial
> and then switch to parallel?
>
> The version is Molpro 2010 mppx.
>
> Thanks in advance for any suggestions.
> -Jan
>
>
> _______________________________________________
> Molpro-user mailing list
> Molpro-user at molpro.net
> http://www.molpro.net/mailman/listinfo/molpro-user