Yes, my point was any testing I do isn't in the form of our usual test
suites. I would like get MTT rolling at some point at LLNL; though I
have (too many :() higher priorities. Also I'm not sure of the 'their
cost' vs. 'our value' ratio of doing runs at much more than 1024 procs
(or even that many).
Andrew
Jeff Squyres wrote:
I think Terry was asking about running at larger scale on a regular
basis for correctness testing (i.e., nightly snapshot tarballs via MTT).
I, for one, would love to see the labs run some of our nightly
tarballs at larger scale so that we have a more consistent datapoint
of what "works" and what "doesn't work" at scale (even if it's a
coarse-grained measurement of the tests we have in ompi-tests). Even
if the tests are not run nightly -- running even a subset of them
even once a week even at "medium" scale would be great. I realize
that even with large clusters, we're all resource-constrained
(needing to let real users run and all that), but any testing on a
regular basis (even if it's sparse) would be really, really great/
useful/good for the code/good for the community/etc.
(yes, this is a not-so-subtle hint :-) )
On Sep 17, 2007, at 11:15 AM, Andrew Friedley wrote:
I won't speak for the labs as a whole, but I generally don't run
things
at scale unless theres something specific I'm after, ie benchmarks or
apps I'm using as a benchmark, rather than test suites.
You might look at some of the purple benchmarks:
http://www.llnl.gov/asci/platforms/purple/rfp/benchmarks/limited/
code_list.html
Andrew
Terry Dontje wrote:
What about Sandia and LANL? Is there anything that is ran on their
large clusters to confirm things seem to work at high np's?
--td
Jeff Squyres wrote:
Cisco is not yet testing that large, but we plan to shortly start
testing at np>=128 (I'm waiting for an internal cluster within Cisco
to be setup properly).
On Sep 11, 2007, at 5:31 PM, rolf.vandeva...@sun.com wrote:
I am curious which tests are being used when running tests on
larger
clusters. And by larger clusters, I mean anything with np > 128.
(Although I realize that is not very large, but it is bigger
than most
of the clusters I assume tests are being run on)
I ask this because I planned on using some of the intel tests, but
they
clearly have limitations starting at np=64.
To avoid mailing list clutter, feel free to just email me and I
will
summarize.
Rolf
_______________________________________________
devel mailing list
de...@open-mpi.org
http://www.open-mpi.org/mailman/listinfo.cgi/devel
_______________________________________________
devel mailing list
de...@open-mpi.org
http://www.open-mpi.org/mailman/listinfo.cgi/devel
_______________________________________________
devel mailing list
de...@open-mpi.org
http://www.open-mpi.org/mailman/listinfo.cgi/devel
