Table of contents
Get it from the SVN. The current version is in the directory named "INRIA". Other directories named "INRIA-*" contain previous versions.
Is there any howto available?
Yes, you can find it in the doc/ section of your QCG-OMPI
directory. Compile it with pdflatex or latex or anything to
build a document from a LaTeX file.
How can I compile / install QCG-OMPI?
Everything is explained in the documentation that comes with
QCG-OMPI, in the doc section.
I cannot find the configure script, where is it?
As explained in the installation documentation ("Howto"), you
need to generate it with autogen.sh.
What software do I need to have installed on my computer?
You will need:
- libpcap and gSOAP for the grid infrastructure
- Autotools (automake, autoconf and libtool) for OpenMPI
- flex or lex, bison or yacc will be asked by gSOAP but
are not needed by QCG-OMPI
Of course all of those programs need to be in your PATH, and the
libraries must be in your LD_LIBRARY_PATH.
Configuration fails with ...
With what? TODO
How do I start my application?
You have to use a deployer that will deploy the grid
infrastructure, setup some parameters and start the MPI
application with the appropriate parameters. Please read the
documentation, everything is written in this not-so-long
document.
mpiexec crashes with a SEGFAULT in OOB TCP (persistent services)
If you get the following error :
i
[qcg:22817] *** Process received signal ***
[qcg:22817] Signal: Segmentation fault (11)
[qcg:22817] Signal code: Address not mapped (1)
[qcg:22817] Failing at address: 0x8
[qcg:22817] [ 0] [0xffffe440]
[qcg:22817] [ 1] /home/inria/persistent/lib/openmpi/mca_oob_tcp.so [0xb7c6593b]
[qcg:22817] [ 2] /home/inria/persistent/lib/openmpi/mca_oob_tcp.so(mca_oob_tcp_peer_resolved+0x43) [0xb7c67b98]
[qcg:22817] [ 3] /home/inria/persistent/lib/openmpi/mca_oob_tcp.so(mca_oob_tcp_resolve+0x54) [0xb7c5e315]
[qcg:22817] [ 4] /home/inria/persistent/lib/openmpi/mca_oob_tcp.so(mca_oob_tcp_peer_send+0x86) [0xb7c637e2]
[qcg:22817] [ 5] /home/inria/persistent/lib/openmpi/mca_oob_tcp.so(mca_oob_tcp_send+0x537) [0xb7c6a627]
[qcg:22817] [ 6] /home/inria/persistent/lib/libopen-rte.so.0(mca_oob_send_packed+0x9b) [0xb7f26e80]
[qcg:22817] [ 7] /home/inria/persistent/lib/openmpi/mca_ns_proxy.so(orte_ns_proxy_create_my_name+0x1e3) [0xb7cacbee]
[qcg:22817] [ 8] /home/inria/persistent/lib/openmpi/mca_sds_singleton.so(orte_sds_singleton_set_name+0x1d) [0xb7c067b1]
[qcg:22817] [ 9] /home/inria/persistent/lib/libopen-rte.so.0(orte_sds_base_set_name+0x37) [0xb7f4fe87]
[qcg:22817] [10] /home/inria/persistent/lib/libopen-rte.so.0(orte_init_stage1+0x488) [0xb7ef6268]
[qcg:22817] [11] /home/inria/persistent/lib/libopen-rte.so.0(orte_system_init+0x24) [0xb7efb2fc]
[qcg:22817] [12] /home/inria/persistent/lib/libopen-rte.so.0(orte_init+0x7b) [0xb7ef5b7f]
[qcg:22817] [13] mpiexec(orterun+0x20f) [0x804a28f]
[qcg:22817] [14] mpiexec(main+0x22) [0x804a076]
[qcg:22817] [15] /lib/i686/cmov/libc.so.6(__libc_start_main+0xe0) [0xb7cf4450]
[qcg:22817] [16] mpiexec [0x8049ff1]
[qcg:22817] *** End of error message ***
Segmentation fault
you may have to restart the grid infrastructure.
Why do I have to change
the range of ports used between two consecutive deployments of the
grid infrastructure?
Because the ports used by the grid infrastructure components in
the previous execution may still be in TIME_WAIT state, that
makes them unusable.
What inter-cluster
communication techniques are implemented, and how can I chose the
one I want to use?
Direct connection, proxy method and TCP traversing. The method
is set in the configuration file of the broker. Use 1 to select
direct connection, 2 to select proxy method and 3 to select TCP
traversing.
How do I specify the
number of processes I want to run on a single machine?
Two solutions:
- Write the name of this machine several times in the
machinefile
- Specify the number of processes running on this machine in
the machinefile: machine slots=nb_of_processes
