On Nov 6, 2014, at 3:39 PM, Joshua Ladd <jladd.m...@gmail.com> wrote:
> Thank you for taking the time to investigate this, Jeff. SC is a hectic and > stressful time for everyone on this list with many deadlines looming. This > bug isn't a priority for us, however, it seems to me that your original > revert, the one that simply wants to disable threading by default (and for > good reason), is a blocker for the 1.8.4 release tomorrow. I would tend to agree. > Therefore, I'm going to once again suggest that unless Nathan finds a > solution by COB today, we live with the error that was made back in 1.8.1 and > punt on this until 1.9. I would still disagree. The performance bug must be fixed. The release needs to be delayed, IMHO. > The current state of the 1.8.4 prerelease is not acceptable with this commit. Yes, I got it. > Thanks to Alina for bringing this issue to light. > > Josh > > > On Thursday, November 6, 2014, Jeff Squyres (jsquyres) <jsquy...@cisco.com> > wrote: > This thread digressed significantly from the original bug report; I did not > realize that the discussion was revolving around the fact that > MPI_THREAD_MULTIPLE no longer works *at all*. > > So here's where we are: > > 1. MPI_THREAD_MULTIPLE doesn't work, even if you --enable-mpi-thread-multiple > > 2. It seems that 2c8087d10b10e0efea176db8907de2720a55454e and > 09b867374e9618007b81bfaf674ec6df04548bed need to be reverted (in that order) > > 3. That restores MPI_THREAD_MULTIPLE functionality (if you > --enable-mpi-thread-multiple) > > 4. However, this brings back the performance problem, too > > I've looked at this all day so far, and am unfortunately just out of time -- > I have some crushing SC deadlines that I *must* meet. :-( > > Nathan will be picking up where I left off later today to see if there's a > simple way to fix just the performance issue for the non-THREAD_MULTIPLE > cases. > > > > On Nov 4, 2014, at 12:15 PM, Alina Sklarevich <ali...@dev.mellanox.co.il> > wrote: > > > Hi, > > > > We observe a hang when running the multi-threading support test "latency.c" > > (attached to this report), which uses MPI_THREAD_MULTIPLE. > > > > The hang happens immediately at the begining of the test and is reproduced > > in the v1.8 release branch. > > > > The command line to reproduce the behavior is: > > > > $ mpirun --map-by node --bind-to core -display-map -np 2 -mca pml ob1 -mca > > btl tcp,self ./thread-tests-1.1/latency > > > > The last commit with which the hang doesn't reproduce is: > > commit: e4d4266d9c69e > > > > > > > > And problems begin after commit: > > > > > > > > commit 09b867374e9618007b81bfaf674ec6df04548bed > > > > Author: Jeff Squyres <jsquy...@cisco.com> > > > > Date: Fri Oct 31 12:42:50 2014 -0700 > > > > > > > > Revert most of open-mpi/ompi@6ef938de3fa9ca0fed2c5bcb0736f65b0d8803af > > > > > > > > Is this expected behavior? In other words, should we not expect any stable > > release in the 1.8.x series to be able to use MPI_THREAD_MULTIPLE with even > > the TCP and SM BTLs? > > > > > > > > Please advise. > > > > > > > > Thanks, > > > > Alina. > > > > > > > > > > > > <latency.c>_______________________________________________ > > devel mailing list > > de...@open-mpi.org > > Subscription: http://www.open-mpi.org/mailman/listinfo.cgi/devel > > Link to this post: > > http://www.open-mpi.org/community/lists/devel/2014/11/16175.php > > > -- > Jeff Squyres > jsquy...@cisco.com > For corporate legal information go to: > http://www.cisco.com/web/about/doing_business/legal/cri/ > > _______________________________________________ > devel mailing list > de...@open-mpi.org > Subscription: http://www.open-mpi.org/mailman/listinfo.cgi/devel > Link to this post: > http://www.open-mpi.org/community/lists/devel/2014/11/16243.php > _______________________________________________ > devel mailing list > de...@open-mpi.org > Subscription: http://www.open-mpi.org/mailman/listinfo.cgi/devel > Link to this post: > http://www.open-mpi.org/community/lists/devel/2014/11/16248.php -- Jeff Squyres jsquy...@cisco.com For corporate legal information go to: http://www.cisco.com/web/about/doing_business/legal/cri/
