[Mtt-users] Using MTT
Greetings all. Andy and I talked the other day about deploying MTT for Open MPI testing. We're still doing lots of active development, but it would be good if we could get you guys to install and run MTT (potentially via cron) and start submitting results back to the database. This is going to be a give-and-take process while active development is still occurring, so bear with us... Here's the general plan: - Everyone needs to send me a username/password to post results (low security stuff, but enough to ensure that random people aren't submitting results to our database). For usernames, let's go by institution: * iu-osl * lanl-acl * hlrs-? (I have no idea what Rainer's group name is :-) ) * utk-icl For a password, use the htpasswd program to create an entry for your username and send me the result (htpasswd is part of apache and is installed on many/most Linux distros). - We'll pick a "stable enough" r number on the mtt trunk in the not- distant future so that everyone can grab this version, make an ini file for your setup, and stuff it in a cron job. - We'll put up a "stable" database at a specific URL somewhere on www.open-mpi.org for your cron job to send its results. - Development will continue on the trunk (and a different database) for all the bugs that we find / features that we'll continue to develop. - We'll probably periodically need to tell everyone to update to a new r number on the trunk (or perhaps we'll use tag branches, or something similar -- we haven't even investigated making tarballs yet). We'll also probably be frequently wiping the "stable" database during this development; we won't be building up much historical data during this process. Hopefully having nightly runs from everyone will *really* help shake out the bugs and feature requests. How does this sound for a plan? -- {+} Jeff Squyres {+} The Open MPI Project {+} http://www.open-mpi.org/
[MTT users] Using mtt For Performance Testing Only
I have a new IB-based cluster. I'm not sure if the IB is running correctly. One way I thought of to test IB is to run MPI tests. mtt looks like it can do this, but it also does a whole lot more that I'm not interested in. So, I'm wondering if it's feasible to run mtt as only a performance testing tool, not as a build test tool. In particular, I was thinking I could modify the ompi-core-perf-testing.ini file to just test a previously installed version of ompi. By the way, I tried to read the archives for this list, which the mailman page says are in http://www.open-mpi.org/MailArchives/mtt-users/ but going there results in a "Not Found" error message. Cordially, -- Jon Forrest Research Computing Support College of Chemistry 173 Tan Hall University of California Berkeley Berkeley, CA 94720-1460 510-643-1032 jlforr...@berkeley.edu
Re: [MTT users] Using mtt For Performance Testing Only
Hi Jon, You can use MTT to do performance testing only. Here is a good starting point for using MTT: https://svn.open-mpi.org/trac/mtt/wiki/OMPITesting Have you tried running trivial tests with your MPI installation? Were you planning on submitting the results to the open-mpi.org database? On Tue, Apr/20/2010 11:12:24AM, Jon Forrest wrote: > I have a new IB-based cluster. I'm not sure > if the IB is running correctly. One way I thought > of to test IB is to run MPI tests. mtt looks > like it can do this, but it also does a whole > lot more that I'm not interested in. > > So, I'm wondering if it's feasible to run > mtt as only a performance testing tool, not > as a build test tool. In particular, I was > thinking I could modify the > ompi-core-perf-testing.ini file to just > test a previously installed version of ompi. > > By the way, I tried to read the archives for > this list, which the mailman page says are > in http://www.open-mpi.org/MailArchives/mtt-users/ > but going there results in a "Not Found" error message. Try this link: http://www.open-mpi.org/community/lists/mtt-users/ Regards, Ethan > > Cordially, > -- > Jon Forrest > Research Computing Support > College of Chemistry > 173 Tan Hall > University of California Berkeley > Berkeley, CA > 94720-1460 > 510-643-1032 > jlforr...@berkeley.edu > ___ > mtt-users mailing list > mtt-us...@open-mpi.org > http://www.open-mpi.org/mailman/listinfo.cgi/mtt-users
Re: [MTT users] Using mtt For Performance Testing Only
On 4/20/2010 2:01 PM, Ethan Mallove wrote: Hi Jon, You can use MTT to do performance testing only. Here is a good starting point for using MTT: https://svn.open-mpi.org/trac/mtt/wiki/OMPITesting Thanks for the pointer! Have you tried running trivial tests with your MPI installation? Yes. I've run various benchmarks and the Infiniband diagnostic programs (not that I always understand their output). But, one of our users is reporting that a standard Chemistry application is not scaling as expected when going to a second node after filling up the first node. This is on a Rocks cluster with 12-cores per node. I figured that doing some more comprehensive MPI tests would be a good place to start. Were you planning on submitting the results to the open-mpi.org database? No. This is just informal testing. -- Jon Forrest Research Computing Support College of Chemistry 173 Tan Hall University of California Berkeley Berkeley, CA 94720-1460 510-643-1032 jlforr...@berkeley.edu
Re: [MTT users] Using mtt For Performance Testing Only
In general, MTT is just an engine for running stuff. We tend to use MTT running lots of MPI correctness and performance tests. So yes, you could use MTT to run your chemistry code over varying numbers of nodes. MTT might be a bit of a big tool for that, but if you have no other infrastructure for this kind of scripting, it may be just the ticket for your needs. On Apr 20, 2010, at 5:10 PM, Jon Forrest wrote: > On 4/20/2010 2:01 PM, Ethan Mallove wrote: > > Hi Jon, > > > > You can use MTT to do performance testing only. Here is a good > > starting point for using MTT: > > > >https://svn.open-mpi.org/trac/mtt/wiki/OMPITesting > > Thanks for the pointer! > > > Have you tried running trivial tests with your MPI installation? > > Yes. I've run various benchmarks and the Infiniband diagnostic > programs (not that I always understand their output). > But, one of our users is reporting that a standard > Chemistry application is not scaling as expected when > going to a second node after filling up the first node. > This is on a Rocks cluster with 12-cores per node. > I figured that doing some more comprehensive MPI tests > would be a good place to start. > > > Were you planning on submitting the results to the open-mpi.org database? > > No. This is just informal testing. > > -- > Jon Forrest > Research Computing Support > College of Chemistry > 173 Tan Hall > University of California Berkeley > Berkeley, CA > 94720-1460 > 510-643-1032 > jlforr...@berkeley.edu > ___ > mtt-users mailing list > mtt-us...@open-mpi.org > http://www.open-mpi.org/mailman/listinfo.cgi/mtt-users > -- Jeff Squyres jsquy...@cisco.com For corporate legal information go to: http://www.cisco.com/web/about/doing_business/legal/cri/
[MTT users] Using MTT to test the newly added SCTP BTL
-BEGIN PGP SIGNED MESSAGE- Hash: SHA1 Hello, I've had some more time to look into running MTT-based tests on the SCTP BTL that has been recently added to ompi-trunk. However, it appears that the BTL itself is not included in the nightly tarballs. I was curious if this is due to the presence of the .ompi_ignore file within the BTL src directory? Any help in this matter would be appreciated. Thanks in advance. - -- Karol Mroz km...@cs.ubc.ca -BEGIN PGP SIGNATURE- Version: GnuPG v1.4.6 (GNU/Linux) iD8DBQFHRcFzuoug78g/Mz8RAnIAAJ98CGeek8JzRF7X+w9B+eO8AhZ4jACfUCop z7hzZ3K83na9uoC4qz0LSMw= =yCE0 -END PGP SIGNATURE-
Re: [MTT users] Using MTT to test the newly added SCTP BTL
Yes, it is. Doh! One solution might be to remove the .ompi_ignore but to only enable the SCTP BTL when an explicit --with-sctp flag is given to configure (or something similar). You might want to run this by the [OMPI] group first, but there's precedent for it, so I doubt anyone would object. On Nov 22, 2007, at 12:50 PM, Karol Mroz wrote: -BEGIN PGP SIGNED MESSAGE- Hash: SHA1 Hello, I've had some more time to look into running MTT-based tests on the SCTP BTL that has been recently added to ompi-trunk. However, it appears that the BTL itself is not included in the nightly tarballs. I was curious if this is due to the presence of the .ompi_ignore file within the BTL src directory? Any help in this matter would be appreciated. Thanks in advance. - -- Karol Mroz km...@cs.ubc.ca -BEGIN PGP SIGNATURE- Version: GnuPG v1.4.6 (GNU/Linux) iD8DBQFHRcFzuoug78g/Mz8RAnIAAJ98CGeek8JzRF7X+w9B+eO8AhZ4jACfUCop z7hzZ3K83na9uoC4qz0LSMw= =yCE0 -END PGP SIGNATURE- ___ mtt-users mailing list mtt-us...@open-mpi.org http://www.open-mpi.org/mailman/listinfo.cgi/mtt-users -- Jeff Squyres Cisco Systems