I decided to try OpenMPI using the 'openib' module, rather than 'mvapi'; however I'm having a bit of difficulty:

The test hardware is the same as in my earlier posts, the only software difference is:

Linux 2.6.14 (OpenIB 2nd gen IB drivers)
OpenIB userspace tools (svn from openib.org)
OpenMPI (svn revision 8046)

I'm executing the program using:

mpirun --prefix /usr/x86_64-gcc-3.3.3/openmpi-1.0svn/ --mca btl openib -np 100 -machinefile nodelist <program>

I receive the following error:
***
[0,1,0][btl_openib_component.c:341:mca_btl_openib_component_init] error obtaining device context for mthca0 errno says No such file or directory

--------------------------------------------------------------------------
No available btl components were found!
***

The output of ompi_info is included; it appears that the openib btl component does exist, however.

Interestingly enough, if I use

mpirun --prefix /usr/x86_64-gcc-3.3.3/openmpi-1.0svn/ --mca ptl openib -np 100 -machinefile nodelist <program>

The program will execute; which is even more interesting:
* There is no openib ptl (or at least, there isn't one in ompi_info, nor is there a corresponding mca_ptl_openib.la or .so file) * Even though 'openib' is specified, the TCP interface is used. (not a bug, but a feature?) * Before execution begins, I receive this error (but execution then continues):
***
[0,1,1][btl_openib_component.c:341:mca_btl_openib_component_init] error obtaining device context for mthca0 errno says No such file or directory
***

Thoughts?

Attachment: ompi_info.out
Description: Binary data

Reply via email to