I've tested a bit more and it looks like this error only occurs when using a process count > 3. That is, mpirun -np 2 or 3 works fine, but with > 4 the error below occurs. This makes me think it is a mpi rather than a serialization issue. Nick On Jun 11, 2009, at 5:00 PM, Matthias Troyer wrote:

Have you tried whether your code van serialize the same vector of pointers into a file and deserialize it successfully again? That test helps decide whether this is an MPI or a serialization issue.

Matthias

On Jun 11, 2009, at 10:13 AM, Nick Collier wrote:

...

Hi,

I'm using boost-1.39.0 with openMpi 1.3.2 on OSX 10.5.7. I'm getting some strange behavior when sending and receiving a vector of pointers between processes. The received pointer addresses are sometimes invalid. I'm not using the skeleton content code, just sending and receiving the vector. Some output:

[2] processing edges from 1 [2] unpacking edge 0x834810 [2] unpacking edge 0x834810 [0] processing edges from 2 [0] unpacking edge 0x834ec0 [0] unpacking edge 0x834ec0 [2] processing edges from 3 [2] unpacking edge 0x8346c0 [2] unpacking edge 0x8346c0 [2] processing edges from 0 [2] unpacking edge 0x1 [2] unpacking edge 0x1

The hex value is the pointer address, and obviously doing anything with the edge with the weird address causes the app to crash. I can't see that I'm doing anything wrong (although of course that's a possibility) and so I thought I'd run the boost mpi tests. The skeleton_content_test fails consistently with the following output:

Broadcasting integer list skeleton from root 0...OK. Broadcasting integer list content from root 0...OK. Broadcasting reversed integer list content from root 0...OK. ../../../boost/test/minimal.hpp(123): exception "memory access violation at address: 0x974bfd1c: non-existent physical address" caught in function: 'int main(int, char**)'

**** Testing aborted.../../../boost/test/minimal.hpp(123): exception "memory access violation at address: 0x974bfd1c: non- existent physical address" caught in function: 'int main(int, char**)'

**** Testing aborted. **** 1 error detected ../../../boost/test/minimal.hpp(123): exception "memory access violation at address: 0x974bfd1c: non-existent physical address" caught in function: 'int main(int, char**)'

**** Testing aborted. **** 1 error detected ../../../boost/test/minimal.hpp(123): exception "memory access violation at address: 0x974bfd1c: non-existent physical address" caught in function: 'int main(int, char**)'

**** Testing aborted. **** 1 error detected ../../../boost/test/minimal.hpp(123): exception "memory access violation at address: 0x974bfd1c: non-existent physical address" caught in function: 'int main(int, char**)'

**** Testing aborted. **** 1 error detected

**** 1 error detected ../../../boost/test/minimal.hpp(123): exception "memory access violation at address: 0x974bfd1c: non-existent physical address" caught in function: 'int main(int, char**)'

**** Testing aborted. **** 1 error detected

I'm not sure if this is related to my problem but the memory issues made me wonder.

Any help is appreciated,

thanks,

Nick

_______________________________________________ Boost-users mailing list Boost-users@lists.boost.org http://lists.boost.org/mailman/listinfo.cgi/boost-users

---

Boost-users mailing list Boost-users@lists.boost.org http://lists.boost.org/mailman/listinfo.cgi/boost-users

What did you ultimately do? Go back to straight MPI, some other work around? I'm having issues with communicator::irecv and communicator::send now as well that I need to diagnose. Nick On Jun 12, 2009, at 3:39 AM, Alain Miniussi wrote:

Hi,

I'm not providing any help but FWIW, I got the same problem with trunk and OpenMpi 1.3.2 on Linux ubuntu 9.04. I also noticed a problem with all_reduce when in_values == out_values, openmpi 1.3.2 detect that MPI_IN_PLACE should be use. I diid not have that one with previous versions of OpenMpi, maybe that release detects more issues than the previous ?

Regards,

On Thu, 11 Jun 2009 18:13:10 +0200, Nick Collier

...

wrote:

...

Hi,

I'm using boost-1.39.0 with openMpi 1.3.2 on OSX 10.5.7. I'm getting some strange behavior when sending and receiving a vector of pointers between processes. The received pointer addresses are sometimes invalid. I'm not using the skeleton content code, just sending and receiving the vector. Some output:

[2] processing edges from 1 [2] unpacking edge 0x834810 [2] unpacking edge 0x834810 [0] processing edges from 2 [0] unpacking edge 0x834ec0 [0] unpacking edge 0x834ec0 [2] processing edges from 3 [2] unpacking edge 0x8346c0 [2] unpacking edge 0x8346c0 [2] processing edges from 0 [2] unpacking edge 0x1 [2] unpacking edge 0x1

The hex value is the pointer address, and obviously doing anything with the edge with the weird address causes the app to crash. I can't see that I'm doing anything wrong (although of course that's a possibility) and so I thought I'd run the boost mpi tests. The skeleton_content_test fails consistently with the following output:

Broadcasting integer list skeleton from root 0...OK. Broadcasting integer list content from root 0...OK. Broadcasting reversed integer list content from root 0...OK. ../../../boost/test/minimal.hpp(123): exception "memory access violation at address: 0x974bfd1c: non-existent physical address" caught in function: 'int main(int, char**)'

**** Testing aborted.../../../boost/test/minimal.hpp(123): exception "memory access violation at address: 0x974bfd1c: non- existent physical address" caught in function: 'int main(int, char**)'

**** Testing aborted. **** 1 error detected ../../../boost/test/minimal.hpp(123): exception "memory access violation at address: 0x974bfd1c: non-existent physical address" caught in function: 'int main(int, char**)'

**** Testing aborted. **** 1 error detected ../../../boost/test/minimal.hpp(123): exception "memory access violation at address: 0x974bfd1c: non-existent physical address" caught in function: 'int main(int, char**)'

**** Testing aborted. **** 1 error detected ../../../boost/test/minimal.hpp(123): exception "memory access violation at address: 0x974bfd1c: non-existent physical address" caught in function: 'int main(int, char**)'

**** Testing aborted. **** 1 error detected

**** 1 error detected ../../../boost/test/minimal.hpp(123): exception "memory access violation at address: 0x974bfd1c: non-existent physical address" caught in function: 'int main(int, char**)'

**** Testing aborted. **** 1 error detected

I'm not sure if this is related to my problem but the memory issues made me wonder.

Any help is appreciated,

thanks,

Nick

_______________________________________________ Boost-users mailing list Boost-users@lists.boost.org http://lists.boost.org/mailman/listinfo.cgi/boost-users

-- Alain Miniussi http://www.oca.eu/alainm Equipe CRIMSON https://crimson.oca.eu +33 492 00 30 09 _______________________________________________ Boost-users mailing list Boost-users@lists.boost.org http://lists.boost.org/mailman/listinfo.cgi/boost-users