[yt-dev] Good benchmark?

Matthew Turk matthewturk at gmail.com
Fri Mar 1 13:34:57 PST 2013


That sounds great.  I have the Santa Fe Light Cone.  I think focusing
on Enzo data (since FLASH data in 2.x has poor performance) would be
best for now.

Would you be able to provide a relatively simple, but compute-heavy,
halo profiling script?  Would the cookbook one work pretty well for
this?  I can handle the other aspects.

On Fri, Mar 1, 2013 at 3:48 PM, Britton Smith <brittonsmith at gmail.com> wrote:
> If we do want to go with some big datasets, I can provide some 1024^3 and
> 1536^3 unigrids.  I'm sure others have better AMR data than I do.
>
>
> On Fri, Mar 1, 2013 at 3:41 PM, Matthew Turk <matthewturk at gmail.com> wrote:
>>
>> Hey Sam,
>>
>> On Fri, Mar 1, 2013 at 2:38 PM, Sam Skillman <samskillman at gmail.com>
>> wrote:
>> > I think in terms of providing a test for the cluster, the benchmark
>> > should
>> > not require the shipment of data.  For the benchmarking of yt for yt's
>> > own
>> > sake, then we do need to test the IO performance.
>>
>> I'm not really sure that we want to avoid IO...
>>
>> >
>> > I think it'd be great to make an in-memory dataset that can then be used
>> > for
>> > testing performance.  For example, set up a bunch of refined spheres and
>> > then project/slice/profile/render/halo find?
>>
>> Perhaps, but will our current IC generators work at really, really big
>> scale?  I didn't think they would.  Plus it's hard to get them to
>> subdivide -- you end up with ltos of nested grids.
>>
>> >
>> > Sam
>> >
>> >
>> > On Fri, Mar 1, 2013 at 12:31 PM, Matthew Turk <matthewturk at gmail.com>
>> > wrote:
>> >>
>> >> Hi Jeff,
>> >>
>> >> Sure.  I got asked for a standard yt performance benchmark to be run
>> >> to figure out performance of an analysis cluster and evaluate its
>> >> readiness.  :)  I think it's safe to sya that we should be pushing
>> >> things like IO, memory capacity and communication performance.
>> >>
>> >> Adding such a set of scripts (and their results!) would be very useful
>> >> going forward, along with a little description of why we chose the
>> >> routines we did.
>> >>
>> >> -Matt
>> >>
>> >> On Fri, Mar 1, 2013 at 2:17 PM, j s oishi <jsoishi at gmail.com> wrote:
>> >> > Matt,
>> >> >
>> >> > It might be helpful if you shared some more details of this
>> >> > particular
>> >> > benchmarking exercise, if you can. It would be helpful for making
>> >> > sure
>> >> > we present the most usefil information to the people asking for it as
>> >> > well as info that is useful to us going forwards.
>> >> >
>> >> > j
>> >> >
>> >> > On Fri, Mar 1, 2013 at 2:05 PM, Matthew Turk <matthewturk at gmail.com>
>> >> > wrote:
>> >> >> Definitely, yes.  For what it's worth, the specific use case I have
>> >> >> in
>> >> >> mind is for performance testing a system, so I think scalability
>> >> >> will
>> >> >> be important.
>> >> >>
>> >> >> On Fri, Mar 1, 2013 at 2:01 PM, Cameron Hummels <chummels at gmail.com>
>> >> >> wrote:
>> >> >>> It might be worth showing how well this works on a single proc as
>> >> >>> well
>> >> >>> as
>> >> >>> how well it works using parallel mode on a few different numbers of
>> >> >>> processors.
>> >> >>>
>> >> >>>
>> >> >>> On Fri, Mar 1, 2013 at 1:57 PM, Christopher Moody
>> >> >>> <chrisemoody at gmail.com>
>> >> >>> wrote:
>> >> >>>>
>> >> >>>> Perhaps a benchmark for simply calculating a derived field? This
>> >> >>>> would
>> >> >>>> then profile a much smaller codebase.
>> >> >>>>
>> >> >>>> Also benchmarking projections/fields for AMR/octree/particles of
>> >> >>>> similar
>> >> >>>> resolutions would be pretty cool.
>> >> >>>>
>> >> >>>> chris
>> >> >>>>
>> >> >>>>
>> >> >>>> On Fri, Mar 1, 2013 at 10:53 AM, Matthew Turk
>> >> >>>> <matthewturk at gmail.com>
>> >> >>>> wrote:
>> >> >>>>>
>> >> >>>>> Okay, I like that idea.  So a unified script with timing for in
>> >> >>>>> each
>> >> >>>>> section might include:
>> >> >>>>>
>> >> >>>>>  * Halo profiling
>> >> >>>>>  * Global projection
>> >> >>>>>  * Global profiles
>> >> >>>>>  * Global VR (CTF and OffAxisProj)
>> >> >>>>>
>> >> >>>>> I think we probably want a large unigrid and a large AMR dataset
>> >> >>>>> to
>> >> >>>>> run these on, too.
>> >> >>>>>
>> >> >>>>> On Fri, Mar 1, 2013 at 1:50 PM, Nathan Goldbaum
>> >> >>>>> <goldbaum at ucolick.org>
>> >> >>>>> wrote:
>> >> >>>>> > Probably a good idea to try some off axis projections or simple
>> >> >>>>> > volume
>> >> >>>>> > renderings to test the VR code.
>> >> >>>>> >
>> >> >>>>> >
>> >> >>>>> > On Fri, Mar 1, 2013 at 10:48 AM, Matthew Turk
>> >> >>>>> > <matthewturk at gmail.com>
>> >> >>>>> > wrote:
>> >> >>>>> >>
>> >> >>>>> >> Hi all,
>> >> >>>>> >>
>> >> >>>>> >> I got an email asking for a benchmark of yt.  I think this is
>> >> >>>>> >> very,
>> >> >>>>> >> very valuable to have going forward.  I was wondering if
>> >> >>>>> >> anyone
>> >> >>>>> >> had
>> >> >>>>> >> any suggestions?
>> >> >>>>> >>
>> >> >>>>> >> I'm thinking that we probably want to test things like the
>> >> >>>>> >> halo
>> >> >>>>> >> finder, projections, and phase plots.  Would a medium (1536^3)
>> >> >>>>> >> halo
>> >> >>>>> >> profiling run do that?  Do we want to add on some global
>> >> >>>>> >> projections
>> >> >>>>> >> and phase plots as well?
>> >> >>>>> >>
>> >> >>>>> >> -Matt
>> >> >>>>> >> _______________________________________________
>> >> >>>>> >> yt-dev mailing list
>> >> >>>>> >> yt-dev at lists.spacepope.org
>> >> >>>>> >> http://lists.spacepope.org/listinfo.cgi/yt-dev-spacepope.org
>> >> >>>>> >
>> >> >>>>> >
>> >> >>>>> >
>> >> >>>>> > _______________________________________________
>> >> >>>>> > yt-dev mailing list
>> >> >>>>> > yt-dev at lists.spacepope.org
>> >> >>>>> > http://lists.spacepope.org/listinfo.cgi/yt-dev-spacepope.org
>> >> >>>>> >
>> >> >>>>> _______________________________________________
>> >> >>>>> yt-dev mailing list
>> >> >>>>> yt-dev at lists.spacepope.org
>> >> >>>>> http://lists.spacepope.org/listinfo.cgi/yt-dev-spacepope.org
>> >> >>>>
>> >> >>>>
>> >> >>>>
>> >> >>>> _______________________________________________
>> >> >>>> yt-dev mailing list
>> >> >>>> yt-dev at lists.spacepope.org
>> >> >>>> http://lists.spacepope.org/listinfo.cgi/yt-dev-spacepope.org
>> >> >>>>
>> >> >>>
>> >> >>>
>> >> >>>
>> >> >>> --
>> >> >>> Cameron Hummels
>> >> >>> Postdoctoral Researcher
>> >> >>> Steward Observatory
>> >> >>> University of Arizona
>> >> >>> http://chummels.org
>> >> >>>
>> >> >>> _______________________________________________
>> >> >>> yt-dev mailing list
>> >> >>> yt-dev at lists.spacepope.org
>> >> >>> http://lists.spacepope.org/listinfo.cgi/yt-dev-spacepope.org
>> >> >>>
>> >> >> _______________________________________________
>> >> >> yt-dev mailing list
>> >> >> yt-dev at lists.spacepope.org
>> >> >> http://lists.spacepope.org/listinfo.cgi/yt-dev-spacepope.org
>> >> > _______________________________________________
>> >> > yt-dev mailing list
>> >> > yt-dev at lists.spacepope.org
>> >> > http://lists.spacepope.org/listinfo.cgi/yt-dev-spacepope.org
>> >> _______________________________________________
>> >> yt-dev mailing list
>> >> yt-dev at lists.spacepope.org
>> >> http://lists.spacepope.org/listinfo.cgi/yt-dev-spacepope.org
>> >
>> >
>> >
>> > _______________________________________________
>> > yt-dev mailing list
>> > yt-dev at lists.spacepope.org
>> > http://lists.spacepope.org/listinfo.cgi/yt-dev-spacepope.org
>> >
>> _______________________________________________
>> yt-dev mailing list
>> yt-dev at lists.spacepope.org
>> http://lists.spacepope.org/listinfo.cgi/yt-dev-spacepope.org
>
>
>
> _______________________________________________
> yt-dev mailing list
> yt-dev at lists.spacepope.org
> http://lists.spacepope.org/listinfo.cgi/yt-dev-spacepope.org
>



More information about the yt-dev mailing list