[MITgcm-support] run speed unaffected by timestep, 'deltaT'

Jody Klymak jklymak at uvic.ca
Tue May 20 12:00:46 EDT 2014


Hi Johnny,

Many queue programs (PBS etc) give you a "walltime" versus "cpu-time" number.  If these are wildly divergent, then you are indeed spending a lot of time on i/o.  You may want to check why that's the case.  For some runs I am doing now I do a full write every 300 time steps or so, and still most of my walltime is cpu time.  Are you writing excessively to STDOUT? 

Cheers,   Jody


On May 20, 2014, at  1:36 AM, Jonny Williams <Jonny.Williams at bristol.ac.uk> wrote:

> Thanks for that Liam
> 
> I'd already a similar suggestion from others about this. I had reduced my output frequency by a factor of 3 (every 10 days to every 30 days) and have been getting a speed up of about a factor of 10. 
> 
> Thanks again!
> 
> Jonny
> 
> 
> On 19 May 2014 17:15, Liam Brannigan <Brannigan at atm.ox.ac.uk> wrote:
> Hi Jonny - are you writing to disk in both runs?  When you increase the time step you may be producing outputs much more frequently.  If writing to disk is a bottleneck on your system then this could greatly reduce the efficiency gain from the larger timestep.
> 
> Liam
> ________________________________________
> From: mitgcm-support-request at mitgcm.org [mitgcm-support-request at mitgcm.org]
> Sent: 19 May 2014 17:00
> To: mitgcm-support at mitgcm.org
> Subject: MITgcm-support Digest, Vol 131, Issue 26
> 
> Send MITgcm-support mailing list submissions to
>         mitgcm-support at mitgcm.org
> 
> To subscribe or unsubscribe via the World Wide Web, visit
>         http://mitgcm.org/mailman/listinfo/mitgcm-support
> or, via email, send a message with subject or body 'help' to
>         mitgcm-support-request at mitgcm.org
> 
> You can reach the person managing the list at
>         mitgcm-support-owner at mitgcm.org
> 
> When replying, please edit your Subject line so it is more specific
> than "Re: Contents of MITgcm-support digest..."
> 
> 
> Today's Topics:
> 
>    1. Re: Problems (Martin Losch)
>    2. run speed unaffected by timestep, 'deltaT', in? (Jonny Williams)
>    3. Re: run speed unaffected by timestep, 'deltaT', in?
>       (Michael Schaferkotter)
>    4. Re: run speed unaffected by timestep, 'deltaT', in?
>       (Jonny Williams)
> 
> 
> ----------------------------------------------------------------------
> 
> Message: 1
> Date: Mon, 19 May 2014 09:46:31 +0200
> From: Martin Losch <Martin.Losch at awi.de>
> To: MITgcm Support <mitgcm-support at mitgcm.org>
> Subject: Re: [MITgcm-support] Problems
> Message-ID: <EEA0E79F-01B2-4528-A369-4CD21C874C3E at awi.de>
> Content-Type: text/plain; charset="windows-1252"
> 
> Hi Mehrdad,
> 
> Michael?s (very good) suggestions with the singleCPUio/globalFile flags unfortunately works only for mds output (the *.meta/data pairs), but for netcdf (i.e., mnc), the singleCPUio/globalFile flags have no effect (not implemented).
> 
> The current workaround is to use post-processing tools. I find the python script utils/python/MITgcmutils/scripts/gluemncbig most useful (requrires a working python installation).
> Alternatively there is the script pair gluemnc xplodemnc in utils/scripts (require the nco operators: nco.sourceforge.net)
> 
> M.
> 
> On May 18, 2014, at 12:40 AM, Michael Schaferkotter <schaferk at bellsouth.net> wrote:
> 
> > For density in output files
> >
> > 1) In ./code sub directory you might have to modify DIAGNOSTICS_SIZE.h, depending on size of domain and how much data you write out.
> >
> > 2) in ./run sub directory
> >
> > a) data.pkg
> >
> >  useDiagnostics=.TRUE.
> >
> >  b) data.diagnostics
> >
> >   &diagnostics_list
> >
> >
> >   frequency(k)=-3600.,                          (For hourly dump)
> >   timePhase(k)=0.,
> >   fields(1,k)='RHOAnoma',
> >    filename(k)='Rhoanoma',
> >
> > Where k is the kth variable in the list.
> >
> >
> >
> >  For combining data into single files,
> >
> > In ./run/data
> >
> > globalFiles=.FALSE.
> > useSingleCpuIO=.TRUE.
> >
> > Also you/ll need to properly configure the mnc package similarly to the diagnostics package and compile with compiler compatible netCDF libraries.
> >
> >
> > Sent from an iOS device.
> >
> >> On May 17, 2014, at 14:48, meh tab <meryzoon at gmail.com> wrote:
> >>
> >> Dear sir/Madam,
> >> I have started using MITgcm software, recently for modeling internal
> >> wave. I have faced some problems and I could not understand the user's
> >> manual guidelines related to these problems. My problems are:
> >> 1- How can we add density variable to nc output file?
> >> 2- When using multiple CPU cores, the results are broken into several
> >> different files. How can I gather the results into one unique file?
> >> Beforehand, I appreciate your kindness for kind attention to my email.
> >> Sincerely Yours,
> >> Mehrdad Tabatabaei
> >>
> >> _______________________________________________
> >> MITgcm-support mailing list
> >> MITgcm-support at mitgcm.org
> >> http://mitgcm.org/mailman/listinfo/mitgcm-support
> >
> > _______________________________________________
> > MITgcm-support mailing list
> > MITgcm-support at mitgcm.org
> > http://mitgcm.org/mailman/listinfo/mitgcm-support
> 
> 
> 
> 
> ------------------------------
> 
> Message: 2
> Date: Mon, 19 May 2014 12:17:18 +0100
> From: Jonny Williams <Jonny.Williams at bristol.ac.uk>
> To: MITgcm Support <mitgcm-support at mitgcm.org>
> Subject: [MITgcm-support] run speed unaffected by timestep, 'deltaT',
>         in?
> Message-ID:
>         <CAA-NaP5GrO+HqD+n90Zob+M3gUM=nveeD6PvNafKZp5YkFzUAQ at mail.gmail.com>
> Content-Type: text/plain; charset="utf-8"
> 
> Hi there
> 
> I am currently running a regional version of the MIgcm with OBCS and EXF
> forcing packages turned on. For the initial few years of the run I am
> running with a very short timestep of 24s and then after that I am running
> with a much linger timestep, 2400s.
> 
> I was expecting a significant increased in run speed but this has not
> happened. I am recompiling in between these run steps.
> 
> Is there something that I've missed?
> 
> Many thanks!
> 
> Jonny
> 
> --
> Dr Jonny Williams
> School of Geographical Sciences
> University of Bristol
> University Road
> BS8 1SS
> 
> +44 (0)117 3318352
> jonny.williams at bristol.ac.uk
> bit.ly/jonnywilliams
> -------------- next part --------------
> An HTML attachment was scrubbed...
> URL: <http://mitgcm.org/pipermail/mitgcm-support/attachments/20140519/7c9cfdc3/attachment.html>
> 
> ------------------------------
> 
> Message: 3
> Date: Mon, 19 May 2014 06:28:59 -0500
> From: Michael Schaferkotter <schaferk at bellsouth.net>
> To: "mitgcm-support at mitgcm.org" <mitgcm-support at mitgcm.org>
> Subject: Re: [MITgcm-support] run speed unaffected by timestep,
>         'deltaT', in?
> Message-ID: <5A22E78C-06EF-4A8F-A3FE-58EB9EB8C393 at bellsouth.net>
> Content-Type: text/plain; charset="us-ascii"
> 
> Since deltaT is a run time parameter, recompiling is unnecessary, though that does not answer your question.
> 
> Sent from an iOS device.
> 
> > On May 19, 2014, at 6:17, Jonny Williams <Jonny.Williams at bristol.ac.uk> wrote:
> >
> > Hi there
> >
> > I am currently running a regional version of the MIgcm with OBCS and EXF forcing packages turned on. For the initial few years of the run I am running with a very short timestep of 24s and then after that I am running with a much linger timestep, 2400s.
> >
> > I was expecting a significant increased in run speed but this has not happened. I am recompiling in between these run steps.
> >
> > Is there something that I've missed?
> >
> > Many thanks!
> >
> > Jonny
> >
> > --
> > Dr Jonny Williams
> > School of Geographical Sciences
> > University of Bristol
> > University Road
> > BS8 1SS
> >
> > +44 (0)117 3318352
> > jonny.williams at bristol.ac.uk
> > bit.ly/jonnywilliams
> > _______________________________________________
> > MITgcm-support mailing list
> > MITgcm-support at mitgcm.org
> > http://mitgcm.org/mailman/listinfo/mitgcm-support
> -------------- next part --------------
> An HTML attachment was scrubbed...
> URL: <http://mitgcm.org/pipermail/mitgcm-support/attachments/20140519/d2729576/attachment-0001.htm>
> 
> ------------------------------
> 
> Message: 4
> Date: Mon, 19 May 2014 12:30:32 +0100
> From: Jonny Williams <Jonny.Williams at bristol.ac.uk>
> To: MITgcm Support <mitgcm-support at mitgcm.org>
> Subject: Re: [MITgcm-support] run speed unaffected by timestep,
>         'deltaT', in?
> Message-ID:
>         <CAA-NaP4QA83X8BkdxxU6tixh-97ob-nsFFj3+UrEsfiBKP0hxQ at mail.gmail.com>
> Content-Type: text/plain; charset="utf-8"
> 
> Hi again
> 
> Regarding this, it seems that I do actually get a speed up of approximately
> 10 times from a factor of 100 difference in the timestep. Is this roughly
> what people would expect?
> 
> Thanks again and sorry for the initial confusion!
> 
> Jonny
> 
> 
> On 19 May 2014 12:17, Jonny Williams <Jonny.Williams at bristol.ac.uk> wrote:
> 
> > Hi there
> >
> > I am currently running a regional version of the MIgcm with OBCS and EXF
> > forcing packages turned on. For the initial few years of the run I am
> > running with a very short timestep of 24s and then after that I am running
> > with a much linger timestep, 2400s.
> >
> > I was expecting a significant increased in run speed but this has not
> > happened. I am recompiling in between these run steps.
> >
> > Is there something that I've missed?
> >
> > Many thanks!
> >
> > Jonny
> >
> > --
> > Dr Jonny Williams
> > School of Geographical Sciences
> > University of Bristol
> > University Road
> > BS8 1SS
> >
> > +44 (0)117 3318352
> > jonny.williams at bristol.ac.uk
> > bit.ly/jonnywilliams
> >
> 
> 
> 
> --
> Dr Jonny Williams
> School of Geographical Sciences
> University of Bristol
> University Road
> BS8 1SS
> 
> +44 (0)117 3318352
> jonny.williams at bristol.ac.uk
> bit.ly/jonnywilliams
> -------------- next part --------------
> An HTML attachment was scrubbed...
> URL: <http://mitgcm.org/pipermail/mitgcm-support/attachments/20140519/f7541a0d/attachment-0001.htm>
> 
> ------------------------------
> 
> _______________________________________________
> MITgcm-support mailing list
> MITgcm-support at mitgcm.org
> http://mitgcm.org/mailman/listinfo/mitgcm-support
> 
> 
> End of MITgcm-support Digest, Vol 131, Issue 26
> ***********************************************
> 
> _______________________________________________
> MITgcm-support mailing list
> MITgcm-support at mitgcm.org
> http://mitgcm.org/mailman/listinfo/mitgcm-support
> 
> 
> 
> -- 
> Dr Jonny Williams
> School of Geographical Sciences
> University of Bristol
> University Road
> BS8 1SS
> 
> +44 (0)117 3318352
> jonny.williams at bristol.ac.uk
> bit.ly/jonnywilliams
> _______________________________________________
> MITgcm-support mailing list
> MITgcm-support at mitgcm.org
> http://mitgcm.org/mailman/listinfo/mitgcm-support

--
Jody Klymak    
http://web.uvic.ca/~jklymak/





-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://mitgcm.org/pipermail/mitgcm-support/attachments/20140520/7f1fdb11/attachment-0001.htm>


More information about the MITgcm-support mailing list