Re: [Apbs-users] Memory usage with mg-para and the usemap keyword

[N. B. <ba...@cc...>]

Hi Scott --

How much memory is available on each node?  How many processes are running on each node?  How big is your grid (dime)?

Thanks,

Nathan

--
Associate Professor, Dept. of Biochemistry and Molecular Biophysics
Center for Computational Biology, Washington University in St. Louis
Web: http://cholla.wustl.edu/

-----Original Message-----
From: Scott Perrin <bs...@ge...>

Date: Thu, 22 May 2008 18:25:25 
To:Nathan Baker <ba...@cc...>
Cc:apb...@li..., Jess Cannata <ja...@ge...>,       "Ainsley A. Gibson" <aa...@ge...>
Subject: Re: [Apbs-users] Memory usage with mg-para and the usemap keyword


Hi,

dime is set to the number of points per processor. If I run the
calculations without reading in the precalculated dx maps, the
calculations completes with no problem. This occurs both with and
without async.

Furthermore, if I were to have set dime to the number of points of the
full grid, I would receive the usual message from APBS that I ran out of
memory. I do not get this message. Instead, APBS will run the node out
of memory and the process will die - along with SSHD.

-Scott


On Thu, 2008-05-22 at 16:37 -0500, Nathan Baker wrote:
> Hello --
>
> Is is possible that you're assuming the "dime" statement in a "mg-
> para" calculation refers to the global dimensions of the calculation?
> In fact, this is the number of grid points per processor (http://apbs.sourceforge.net/doc/user-guide/index.html#elec-dime
> ) and should be adjusted to fit the memory available for each
> processor.  In general, you can assume that APBS will use
> approximately 160 bytes per grid point.  Therefore, you'll want to set
> dime to use an appropriate number of grid points on each processor.
>
> I think this effectively answers your questions below.  Please let me
> know if you still have questions.
>
> -- Nathan
>
> On May 22, 2008, at 10:10 AM, Bradley Scott Perrin wrote:
>
> > Dear all,
> >
> > I have a few questions with regards to memory usage for a mg-para
> > calculation including the usemap keyword.
> >
> > 1. Can memory usage be estimated for a run when dx maps are used for
> > the dielectric grid? I believe the memory requirement is greater
> > than that calculated by psize.py.
> >
> > 2. Is the memory distributed evenly over all processors? If not, how
> > is it distributed? Can I predict the mamximum memory used by one
> > processor?
> >
> > 3. Can I restrict the amount of memory used by APBS per node and per
> > process?
> >
> > Essentially, I am having an issue where a parallel APBS calculation
> > will max out the memory on a node leading to SSHD shutting down. To
> > prevent further aggrivation of my system administrator it would be
> > nice to avoid this problem.
> >
> > Thank you in advance,
> > Scott Perrin
> >
> >
> > -------------------------------------------------------------------------
> > This SF.net email is sponsored by: Microsoft
> > Defy all challenges. Microsoft(R) Visual Studio 2008.
> > http://clk.atdmt.com/MRT/go/vse0120000070mrt/direct/01/
> > _______________________________________________
> > apbs-users mailing list
> > apb...@li...
> > https://lists.sourceforge.net/lists/listinfo/apbs-users
>
> --
> Associate Professor, Dept. of Biochemistry and Molecular Biophysics
> Center for Computational Biology, Washington University in St. Louis
> Web: http://cholla.wustl.edu/
>
>
>
>
>