Re: Re-starting a calculation

["Angel J. Garcia-Adeva" <angarcia@lanl.gov>]

Hi Paula,

There is a small checkpointing utility called chpox that can be
downloaded from sourceforge.net. It allows saving the state of any
process in a file and restarting it later from that point. In conjuction
with a small shell script and cron it is quite powerful. I know that it
works well for intel processors. The author claims that it also works
with alpha architecture, but I have had problems compiling it. Please,
let me know if this helps.

Angel,

PS, maybe this module could be integrated with FEFF in a future release?


On Mon, 2002-12-02 at 17:56, John J. Rehr wrote:
> Paula Harkins asked about restarting a FEFF8 calculation:
> 
> >I am using Feff8 (ver 8.1) and wondered if there was any way to restart a 
> >calculation from the point that it was stopped at rather than having to 
> >wait for the whole calculation to run again. (I share my computer with 
> >several other users!) Any ideas?
> 
>   At present this can only be done module by module. For example if a 
> calculation stops during  the 3rd module [FMS], it is not necessary to
> rerun the first two modules and one can restart the calculation with
> 
> CONTROL 0 0 1 1 1 1
> 
> You can check that a module has finished by looking at the last line
> in logn.dat. For example, when the potentials have been calculated the
> last line should say:
> 
> Done with module 1: potentials.
> 
>   What else can be done?
> 
>   If you do want to stop a job temporarily, you can use the ctrl-z command
> in unix to stop a job and then restart later with the command fg.
> 
> 
>   Several things can be done to optimize the speed to avoid the need for
> stopping/restarting:
> 
> i)  For XANES calculations, the time goes up as [N(lmax+1)**2]**3,
> so you should keep the cluster size as small as you can get away
> with without sacrificing accuracy [typically 30-40 atoms] and
> with default lmax [set lmax = -1 in POTENTIALS].
> 
> ii)  For the FMS part, if you have FEFF8.2 the calculations go faster with the
> Lanczos algorithm.
> 
>   We'll look into the possibility of restarting a FEFF8 calculation mid 
> module. Certainly some band structure codes allow one to restart during
> the SCF loop.
> 
>    Cheers,
>    John Rehr
> 
> 
> 
>