[Gfs-devel] bug with gfs_domain_tag_droplets

20 views
Skip to first unread message

Daniel Fuster

unread,
Oct 8, 2012, 7:23:20 PM10/8/12
to GFS developper discussion list
Hi Stephane

I have some problems using GfsOutputDropletSums in parallel for a very
large number of droplets (error below). After some time digging into
the problem I found that indeed the subroutine reduce_touching_regions
tries to fill out the output array in a component larger than the
allocated size.

The problem seems to be solved if I swith off the commutive flag. That
is, I replace
MPI_Op_create (reduce_touching_regions, TRUE, &op);
by
MPI_Op_create (reduce_touching_regions, FALSE, &op);

The pdf seems to be correct.

Could you confirm that I am allowed to do this?

Thanks
Daniel


[bazant:24649] *** Process received signal ***
[bazant:24649] Signal: Segmentation fault (11)
[bazant:24649] Signal code: Address not mapped (1)
[bazant:24649] Failing at address: 0x10647928
gerris3D: malloc.c:3097: sYSMALLOc: Assertion `(old_top == (((mbinptr)
(((char *) &((av)->bins[((1) - 1) * 2])) - __builtin_offsetof (struct
malloc_chunk, fd)))) && old_size == 0) || ((unsigned long) (old_size)
>= (unsigned long)((((__builtin_offsetof (struct malloc_chunk,
fd_nextsize))+((2 * (sizeof(size_t))) - 1)) & ~((2 * (sizeof(size_t)))
- 1))) && ((old_top)->size & 0x1) && ((unsigned long)old_end &
pagemask) == 0)' failed.
[bazant:24649] *** Process received signal ***
[bazant:24649] Signal: Aborted (6)
[bazant:24649] Signal code: (-6)
[bazant:24649] [ 0] /lib/libpthread.so.0(+0xeff0) [0x7f5cb2068ff0]
[bazant:24649] [ 1] /lib/libc.so.6(gsignal+0x35) [0x7f5cb1d2a1b5]
[bazant:24649] [ 2] /lib/libc.so.6(abort+0x180) [0x7f5cb1d2cfc0]
[bazant:24649] [ 3] /lib/libc.so.6(+0x7201a) [0x7f5cb1d6a01a]
[bazant:24649] [ 4] /lib/libc.so.6(+0x74f6d) [0x7f5cb1d6cf6d]
[bazant:24649] [ 5] /lib/libc.so.6(__libc_malloc+0x70) [0x7f5cb1d6ec70]
[bazant:24649] [ 6] /lib/libc.so.6(__backtrace_symbols+0x11f) [0x7f5cb1ddc58f]
[bazant:24649] [ 7]
/usr/lib/libopen-pal.so.0(opal_backtrace_buffer+0x3a) [0x7f5cb28c963a]
[bazant:24649] [ 8] /usr/lib/libopen-pal.so.0(+0x3218c) [0x7f5cb28c718c]
[bazant:24649] [ 9] /lib/libpthread.so.0(+0xeff0) [0x7f5cb2068ff0]
[bazant:24649] [10] /home/fuster/lib/libgfs3D-1.3.so.2(+0x4ec67)
[0x7f5cb3dffc67]
[bazant:24649] [11] /home/fuster/lib/libgfs3D-1.3.so.2(+0x4ef73)
[0x7f5cb3dfff73]
[bazant:24649] [12]
/usr/lib/openmpi/lib/openmpi/mca_coll_tuned.so(+0x5061)
[0x7f5cacf32061]
[bazant:24649] [13] /usr/lib/libmpi.so.0(PMPI_Allreduce+0x193) [0x7f5cb2d82bd3]
[bazant:24649] [14]
/home/fuster/lib/libgfs3D-1.3.so.2(gfs_domain_tag_droplets+0x1f1)
[0x7f5cb3dff791]
[bazant:24649] [15] /home/fuster/lib/libgfs3D-1.3.so.2(+0x3093f)
[0x7f5cb3de193f]
[bazant:24649] [16]
/home/fuster/lib/libgfs3D-1.3.so.2(gfs_event_do+0x55) [0x7f5cb3e0cdc5]

--
www.danielfuster.com

------------------------------------------------------------------------------
Don't let slow site performance ruin your business. Deploy New Relic APM
Deploy New Relic app performance management and know exactly
what is happening inside your Ruby, Python, PHP, Java, and .NET app
Try New Relic at no cost today and get our sweet Data Nerd shirt too!
http://p.sf.net/sfu/newrelic-dev2dev
_______________________________________________
Gfs-devel mailing list
Gfs-...@lists.sourceforge.net
https://lists.sourceforge.net/lists/listinfo/gfs-devel

gilou agbaglah

unread,
Oct 23, 2012, 2:50:28 PM10/23/12
to gfs-...@lists.sourceforge.net
Hi Daniel,
Thanks a lot...Your trick solved the bug in my system.

Best

2012/10/12 gilou agbaglah <gilou.a...@gmail.com>
Ok, thanks you...I will have a look to that.


2012/10/12 Daniel Fuster <dfu...@gmail.com>
Here an email from me to the devel list never replied from Dr. SP. It
may be of interest for you :)

You will need to change the code, recompile and see if that fixes the
problem. If that is the case, you can send an email to the mailing
list saying so :)

cu
Daniel

Stephane Popinet

unread,
Nov 13, 2012, 7:04:08 PM11/13/12
to GFS developper discussion list
Hi Daniel,

> The problem seems to be solved if I swith off the commutive flag. That
> is, I replace
> MPI_Op_create (reduce_touching_regions, TRUE, &op);
> by
> MPI_Op_create (reduce_touching_regions, FALSE, &op);
>
> The pdf seems to be correct.
>
> Could you confirm that I am allowed to do this?

Hmm, I am not sure I fully understand what is going on there. I
suspect that your fix does not really fix the underlying problem i.e.
it may fail again for another problem. Do you have a parameter file
which reproduces this?

I will add your fix in the next release but it would be good to get to
the bottom of it.

cheers

Stephane

------------------------------------------------------------------------------
Monitor your physical, virtual and cloud infrastructure from a single
web console. Get in-depth insight into apps, servers, databases, vmware,
SAP, cloud infrastructure, etc. Download 30-day Free Trial.
Pricing starts from $795 for 25 servers or applications!
http://p.sf.net/sfu/zoho_dev2dev_nov
Reply all
Reply to author
Forward
0 new messages