myricom MX udapl problems with intel MPI

myricom MX udapl problems with intel MPI

Hi,

We have intel cluster toolkit compiler edition 3.2 installed and we are trying to use the intel mpi libraries in ICT to run on Myricom's MX.

I have downloaded the udapl library from from source forge and installed the DAPL 2.0 version in package dapl-myrinet-20-0.4.tar.gz.

when I try to run mpi this is what i get.
hpcp7761(salmr0)61:mpiexec -perhost 1 -genv I_MPI_DEVICE rdma:mx -np 2 -env I_MPI_DEBUG 10 ../bin/x86_64/IMB-MPI1.intelmpi PingPong
[1] MPI startup(): cannot open dynamic library libdat2.so
[1] my_dlopen(): Look for library libdat2.so in /usr/local/udaplmx/lib64/,/hpc/soft/intel/x86_64/ict-3.2.020/impi/3.2.0.011/lib64,/hpc/soft/intel/x86_64/ict-3.2.020/itac/7.2.0.011/itac/slib_impi3,/hpc/soft/intel/x86_64/ict-3.2.020/fc/11.0.074/lib/intel64,/hpc/soft/intel/x86_64/ict-3.2.020/cc/11.0.074/lib/intel64,/hpc/soft/intel/x86_64/ict-3.2.020/mkl/10.1.0.015/lib/em64t/usr/X11R6/lib64/Xaw3d,/usr/X11R6/lib64,/usr/X11R6/lib/Xaw3d,/usr/X11R6/lib,/usr/x86_64-suse-linux/lib,/usr/local/lib,/opt/kde3/lib,/opt/gnome/lib,/lib64,/lib,/usr/lib64,/usr/lib,/usr/local/lib64,/opt/kde3/lib64,/opt/gnome/lib64,include /etc/ld.so.conf.d/*.conf,/usr/local/syscfg/lib,,/lib,/usr/lib
[1] MPI startup(): cannot open dynamic library libdat2.so

hpcp7761(salmr0)59:ls /usr/local/udaplmx/lib64/
libdapl.a libdapl.so libdat2.so libdat.a libdat.so

hpcp7761(salmr0)65:cat /etc/dat.conf
#
# DAT 2.0 configuration file
#
# Each entry should have the following fields:
#
# \
#
#
mx u2.0 nonthreadsafe default /usr/local/udaplmx/lib64/libdapl.so ofa.2.0 "mx0 0" ""

Any ideas as to what the problem is?

Thanks
Rene

4 posts / 0 new
Last post
For more complete information about compiler optimizations, see our Optimization Notice.

I'm not an Intel MPI expert, but I speculate that the problem is either the "/" on the end of "/usr/local/udaplmx/lib64/" in your search path, or /usr/local/udaplmx/lib64 is not available on a remote compute node, and the message is coming from there. If not, I bet help@myri.com can help.

--Glenn Brown (a Myricom developer)

Quoting - g23754

I'm not an Intel MPI expert, but I speculate that the problem is either the "/" on the end of "/usr/local/udaplmx/lib64/" in your search path, or /usr/local/udaplmx/lib64 is not available on a remote compute node, and the message is coming from there. If not, I bet help@myri.com can help.

--Glenn Brown (a Myricom developer)

According to the message, /usr/local/udaplmx/lib64 wasn't included in the search path LD_LIBRARY_PATH, on the node where the search failed. Intel MPI couldn't be expected to add (or subtract) that directory from the search path.

Quoting - tim18

According to the message, /usr/local/udaplmx/lib64 wasn't included in the search path LD_LIBRARY_PATH, on the node where the search failed. Intel MPI couldn't be expected to add (or subtract) that directory from the search path.

Yep that was it. Thanks. I added setenv LD_LIBRARY_PATH /usr/local/udaplmx/lib64:${LD_LIBRARY_PATH}
to my .cshrc file and now things work. Which is kind of strange because I have

cat /etc/ld.so.conf.d/udaplmx.ld.conf
/usr/local/udaplmx/lib64

setup system wide so there should be no need for me to set LD_LIBRARY_PATH.

Thanks

Rene

Login to leave a comment.