openstack team mailing list archive

[Reza Bakhshayeshi <reza.b2008@xxxxxxxxx>]

Hi,

I have installed OpenStack Folsom on Ubuntu and I want to run HPCC
benchmark on that.
For this reason I have created an image with
OpenMPI 1.6.4 and
Atlas 3.10.1 and
HPCC 1.4.2
I can run the test on single instance and from one instance on another. But
the problem arise when I want to run the test simultaneously on two
instances with:
mpirun -np 2 --hostfile myhosts hpcc
then I receive the following error:

[ubuntu-benchmark:03607] *** Process received signal ***
[ubuntu-benchmark:03607] Signal: Segmentation fault (11)
[ubuntu-benchmark:03607] Signal code: Address not mapped (1)
[ubuntu-benchmark:03607] Failing at address: 0x26ea000
[ubuntu-benchmark:03607] [ 0] /lib/x86_64-linux-gnu/libc.so.6(+0x364a0)
[0x2b92c5ba54a0]
[ubuntu-benchmark:03607] [ 1]
hpcc(HPCC_Power2NodesMPIRandomAccessCheck_LCG+0xa46) [0x4531a6]
[ubuntu-benchmark:03607] [ 2] hpcc(HPCC_MPIRandomAccess_LCG+0x87a)
[0x427b3a]
[ubuntu-benchmark:03607] [ 3] hpcc(main+0xd1b) [0x409e1b]
[ubuntu-benchmark:03607] [ 4]
/lib/x86_64-linux-gnu/libc.so.6(__libc_start_main+0xed) [0x2b92c5b9076d]
[ubuntu-benchmark:03607] [ 5] hpcc() [0x40a9fd]
[ubuntu-benchmark:03607] *** End of error message ***
[ubuntu-benchmark][[17032,1],0][btl_tcp_frag.c:215:mca_btl_tcp_frag_recv]
mca_btl_tcp_frag_recv: readv failed: Connection reset by peer (104)
--------------------------------------------------------------------------
mpirun noticed that process rank 1 with PID 3607 on node ubuntu-benchmark02
exited on signal 11 (Segmentation fault).
--------------------------------------------------------------------------

I have tested different scenarios (for example installing openmpi or mpich2
via apt-get) but every time I still receive the same error. Now I think one
possible problem source could be from OpenStack itself. I don't know maybe
memory allocation strategy (in worse case can openstack fake the allocated
memory?) or something like that (I have tested both KVM and QEMU).
Any idea would be helpful for me.
Thanks for your consideration.

Regards,
Reza