yt/mpi4py on oak ridge cray machines
Hi all, I’ve been trying to build yt and mpi4py on the cray machines at oak ridge and came across this old thread in the mailing list: http://lists.spacepope.org/pipermail/yt-users-spacepope.org/2013-December/01... I was just wondering if anyone had managed this since? If so, what the best approach is (system python or install script)? Many thanks, David
Hi David, I've recently been using Titan and have found it more efficient to build my own mpi4py and python, with the install script. -Matt On Wed, Aug 13, 2014 at 9:16 AM, David Sullivan <D.Sullivan@sussex.ac.uk> wrote:
Hi all,
I’ve been trying to build yt and mpi4py on the cray machines at oak ridge and came across this old thread in the mailing list: http://lists.spacepope.org/pipermail/yt-users-spacepope.org/2013-December/01...
I was just wondering if anyone had managed this since? If so, what the best approach is (system python or install script)?
Many thanks, David
_______________________________________________ yt-users mailing list yt-users@lists.spacepope.org http://lists.spacepope.org/listinfo.cgi/yt-users-spacepope.org
Matt, now that 3.0 is out, perhaps it's time to start contacting the various centers and push for a central install of yt? On Wed, Aug 13, 2014 at 11:21 AM, Matthew Turk <matthewturk@gmail.com> wrote:
Hi David,
I've recently been using Titan and have found it more efficient to build my own mpi4py and python, with the install script.
-Matt
On Wed, Aug 13, 2014 at 9:16 AM, David Sullivan <D.Sullivan@sussex.ac.uk> wrote:
Hi all,
I’ve been trying to build yt and mpi4py on the cray machines at oak ridge and came across this old thread in the mailing list:
http://lists.spacepope.org/pipermail/yt-users-spacepope.org/2013-December/01...
I was just wondering if anyone had managed this since? If so, what the
best
approach is (system python or install script)?
Many thanks, David
_______________________________________________ yt-users mailing list yt-users@lists.spacepope.org http://lists.spacepope.org/listinfo.cgi/yt-users-spacepope.org
_______________________________________________ yt-users mailing list yt-users@lists.spacepope.org http://lists.spacepope.org/listinfo.cgi/yt-users-spacepope.org
-- Michael Zingale Associate Professor Dept. of Physics & Astronomy • Stony Brook University • Stony Brook, NY 11794-3800 *phone*: 631-632-8225 *e-mail*: Michael.Zingale@stonybrook.edu *web*: http://www.astro.sunysb.edu/mzingale
On Wed, Aug 13, 2014 at 10:29 AM, Michael Zingale <michael.zingale@stonybrook.edu> wrote:
Matt, now that 3.0 is out, perhaps it's time to start contacting the various centers and push for a central install of yt?
Yes, absolutely. We may need to put out a minor bugfix release in a few weeks, but I don't see any reason we can't start the ball rolling now. Perhaps: * Blue Waters (I'll take this one) * Stampede * Titan * Nautilus * Gordon * NERSC Any others? As a bit of additional info, one side effect of the PNaCl porting is that we now have a repeatable static build of python. At big supercomputer centers with parallel file systems, the time to import a parallel python job tends to be dominated by the cost of the calls to the metadata nodes, searching for .py files. With the PNaCl build, Mike, Sam and I managed to get a zipimporter working that will import any python files from a single central file (i.e., same amount of data, vastly fewer calls to the MDS) and use the built-in static libraries. Right now I'm stuck on the fact that the glibc on Titan is a billion years old, but my semi-optimistic hope is that eventually we'll have a single binary loader with no external dependencies that requires only a handful of files to import. -Matt
On Wed, Aug 13, 2014 at 11:21 AM, Matthew Turk <matthewturk@gmail.com> wrote:
Hi David,
I've recently been using Titan and have found it more efficient to build my own mpi4py and python, with the install script.
-Matt
On Wed, Aug 13, 2014 at 9:16 AM, David Sullivan <D.Sullivan@sussex.ac.uk> wrote:
Hi all,
I’ve been trying to build yt and mpi4py on the cray machines at oak ridge and came across this old thread in the mailing list:
http://lists.spacepope.org/pipermail/yt-users-spacepope.org/2013-December/01...
I was just wondering if anyone had managed this since? If so, what the best approach is (system python or install script)?
Many thanks, David
_______________________________________________ yt-users mailing list yt-users@lists.spacepope.org http://lists.spacepope.org/listinfo.cgi/yt-users-spacepope.org
_______________________________________________ yt-users mailing list yt-users@lists.spacepope.org http://lists.spacepope.org/listinfo.cgi/yt-users-spacepope.org
-- Michael Zingale Associate Professor
Dept. of Physics & Astronomy • Stony Brook University • Stony Brook, NY 11794-3800 phone: 631-632-8225 e-mail: Michael.Zingale@stonybrook.edu web: http://www.astro.sunysb.edu/mzingale
_______________________________________________ yt-users mailing list yt-users@lists.spacepope.org http://lists.spacepope.org/listinfo.cgi/yt-users-spacepope.org
* Blacklight On Wed, Aug 13, 2014 at 4:37 PM, Matthew Turk <matthewturk@gmail.com> wrote:
On Wed, Aug 13, 2014 at 10:29 AM, Michael Zingale <michael.zingale@stonybrook.edu> wrote:
Matt, now that 3.0 is out, perhaps it's time to start contacting the various centers and push for a central install of yt?
Yes, absolutely. We may need to put out a minor bugfix release in a few weeks, but I don't see any reason we can't start the ball rolling now. Perhaps:
* Blue Waters (I'll take this one) * Stampede * Titan * Nautilus * Gordon * NERSC
Any others?
As a bit of additional info, one side effect of the PNaCl porting is that we now have a repeatable static build of python. At big supercomputer centers with parallel file systems, the time to import a parallel python job tends to be dominated by the cost of the calls to the metadata nodes, searching for .py files. With the PNaCl build, Mike, Sam and I managed to get a zipimporter working that will import any python files from a single central file (i.e., same amount of data, vastly fewer calls to the MDS) and use the built-in static libraries. Right now I'm stuck on the fact that the glibc on Titan is a billion years old, but my semi-optimistic hope is that eventually we'll have a single binary loader with no external dependencies that requires only a handful of files to import.
-Matt
On Wed, Aug 13, 2014 at 11:21 AM, Matthew Turk <matthewturk@gmail.com> wrote:
Hi David,
I've recently been using Titan and have found it more efficient to build my own mpi4py and python, with the install script.
-Matt
On Wed, Aug 13, 2014 at 9:16 AM, David Sullivan <
D.Sullivan@sussex.ac.uk>
wrote:
Hi all,
I’ve been trying to build yt and mpi4py on the cray machines at oak ridge and came across this old thread in the mailing list:
http://lists.spacepope.org/pipermail/yt-users-spacepope.org/2013-December/01...
I was just wondering if anyone had managed this since? If so, what the best approach is (system python or install script)?
Many thanks, David
_______________________________________________ yt-users mailing list yt-users@lists.spacepope.org http://lists.spacepope.org/listinfo.cgi/yt-users-spacepope.org
_______________________________________________ yt-users mailing list yt-users@lists.spacepope.org http://lists.spacepope.org/listinfo.cgi/yt-users-spacepope.org
-- Michael Zingale Associate Professor
Dept. of Physics & Astronomy • Stony Brook University • Stony Brook, NY 11794-3800 phone: 631-632-8225 e-mail: Michael.Zingale@stonybrook.edu web: http://www.astro.sunysb.edu/mzingale
_______________________________________________ yt-users mailing list yt-users@lists.spacepope.org http://lists.spacepope.org/listinfo.cgi/yt-users-spacepope.org
_______________________________________________ yt-users mailing list yt-users@lists.spacepope.org http://lists.spacepope.org/listinfo.cgi/yt-users-spacepope.org
Hi, I work in the user services group at NERSC, and have installed yt 2.6.1 on all the systems as a module. I was planning to do 3.0 when I have a chance. However, that's would just be the standard install which would have all the problems with parallel python jobs that Matt mentions. If there is an improved version, I'd be willing to work with someone to help get it installed. Bye, Lisa On 8/13/14 8:37 AM, Matthew Turk wrote:
On Wed, Aug 13, 2014 at 10:29 AM, Michael Zingale <michael.zingale@stonybrook.edu> wrote:
Matt, now that 3.0 is out, perhaps it's time to start contacting the various centers and push for a central install of yt?
Yes, absolutely. We may need to put out a minor bugfix release in a few weeks, but I don't see any reason we can't start the ball rolling now. Perhaps:
* Blue Waters (I'll take this one) * Stampede * Titan * Nautilus * Gordon * NERSC
Any others?
As a bit of additional info, one side effect of the PNaCl porting is that we now have a repeatable static build of python. At big supercomputer centers with parallel file systems, the time to import a parallel python job tends to be dominated by the cost of the calls to the metadata nodes, searching for .py files. With the PNaCl build, Mike, Sam and I managed to get a zipimporter working that will import any python files from a single central file (i.e., same amount of data, vastly fewer calls to the MDS) and use the built-in static libraries. Right now I'm stuck on the fact that the glibc on Titan is a billion years old, but my semi-optimistic hope is that eventually we'll have a single binary loader with no external dependencies that requires only a handful of files to import.
-Matt
On Wed, Aug 13, 2014 at 11:21 AM, Matthew Turk <matthewturk@gmail.com> wrote:
Hi David,
I've recently been using Titan and have found it more efficient to build my own mpi4py and python, with the install script.
-Matt
On Wed, Aug 13, 2014 at 9:16 AM, David Sullivan <D.Sullivan@sussex.ac.uk> wrote:
Hi all,
I’ve been trying to build yt and mpi4py on the cray machines at oak ridge and came across this old thread in the mailing list:
http://lists.spacepope.org/pipermail/yt-users-spacepope.org/2013-December/01...
I was just wondering if anyone had managed this since? If so, what the best approach is (system python or install script)?
Many thanks, David
_______________________________________________ yt-users mailing list yt-users@lists.spacepope.org http://lists.spacepope.org/listinfo.cgi/yt-users-spacepope.org
_______________________________________________ yt-users mailing list yt-users@lists.spacepope.org http://lists.spacepope.org/listinfo.cgi/yt-users-spacepope.org
-- Michael Zingale Associate Professor
Dept. of Physics & Astronomy • Stony Brook University • Stony Brook, NY 11794-3800 phone: 631-632-8225 e-mail: Michael.Zingale@stonybrook.edu web: http://www.astro.sunysb.edu/mzingale
_______________________________________________ yt-users mailing list yt-users@lists.spacepope.org http://lists.spacepope.org/listinfo.cgi/yt-users-spacepope.org
_______________________________________________ yt-users mailing list yt-users@lists.spacepope.org http://lists.spacepope.org/listinfo.cgi/yt-users-spacepope.org
*Pleiades and Discover On Wednesday, August 13, 2014, Matthew Turk <matthewturk@gmail.com> wrote:
On Wed, Aug 13, 2014 at 10:29 AM, Michael Zingale <michael.zingale@stonybrook.edu <javascript:;>> wrote:
Matt, now that 3.0 is out, perhaps it's time to start contacting the various centers and push for a central install of yt?
Yes, absolutely. We may need to put out a minor bugfix release in a few weeks, but I don't see any reason we can't start the ball rolling now. Perhaps:
* Blue Waters (I'll take this one) * Stampede * Titan * Nautilus * Gordon * NERSC
Any others?
As a bit of additional info, one side effect of the PNaCl porting is that we now have a repeatable static build of python. At big supercomputer centers with parallel file systems, the time to import a parallel python job tends to be dominated by the cost of the calls to the metadata nodes, searching for .py files. With the PNaCl build, Mike, Sam and I managed to get a zipimporter working that will import any python files from a single central file (i.e., same amount of data, vastly fewer calls to the MDS) and use the built-in static libraries. Right now I'm stuck on the fact that the glibc on Titan is a billion years old, but my semi-optimistic hope is that eventually we'll have a single binary loader with no external dependencies that requires only a handful of files to import.
-Matt
On Wed, Aug 13, 2014 at 11:21 AM, Matthew Turk <matthewturk@gmail.com
<javascript:;>>
wrote:
Hi David,
I've recently been using Titan and have found it more efficient to build my own mpi4py and python, with the install script.
-Matt
On Wed, Aug 13, 2014 at 9:16 AM, David Sullivan <
D.Sullivan@sussex.ac.uk <javascript:;>>
wrote:
Hi all,
I’ve been trying to build yt and mpi4py on the cray machines at oak ridge and came across this old thread in the mailing list:
http://lists.spacepope.org/pipermail/yt-users-spacepope.org/2013-December/01...
I was just wondering if anyone had managed this since? If so, what the best approach is (system python or install script)?
Many thanks, David
_______________________________________________ yt-users mailing list yt-users@lists.spacepope.org <javascript:;> http://lists.spacepope.org/listinfo.cgi/yt-users-spacepope.org
_______________________________________________ yt-users mailing list yt-users@lists.spacepope.org <javascript:;> http://lists.spacepope.org/listinfo.cgi/yt-users-spacepope.org
-- Michael Zingale Associate Professor
Dept. of Physics & Astronomy • Stony Brook University • Stony Brook, NY 11794-3800 phone: 631-632-8225 e-mail: Michael.Zingale@stonybrook.edu <javascript:;> web: http://www.astro.sunysb.edu/mzingale
_______________________________________________ yt-users mailing list yt-users@lists.spacepope.org <javascript:;> http://lists.spacepope.org/listinfo.cgi/yt-users-spacepope.org
_______________________________________________ yt-users mailing list yt-users@lists.spacepope.org <javascript:;> http://lists.spacepope.org/listinfo.cgi/yt-users-spacepope.org
participants (6)
-
Britton Smith -
David Sullivan -
Lisa Gerhardt -
Matthew Turk -
Michael Zingale -
Nathan Goldbaum