[gridengine users] Unable to find pe_start file

[Ursula Winkler]

Reuti wrote:
> Hi,
>
> Am 03.04.2012 um 10:48 schrieb Ursula Winkler:
>
>   
>> Dear members of gridengine,
>>
>> I have problems running mpi jobs with SGE. The error message is:
>>                                           "unable to find pe_start file "/softw".
>>
>> Unfortunately not the whole path is printed so I can just guess.
>>
>> I created with qmon --> Parallel Environment Configuration the Environment: "mpi" -->
>> Start Proc Args: /software/sge/mpi/startmpi.sh -catch_rsh $pe_hostfile
>> Stop Proc Args: /software/sge/mpi/stopmpi.sh
>>     
>
> which MPI implementation do you intend to use? With Open MPI and MPICH2 you get a Tight Integration nowadays without and special procedures.
>
>   
I guess the tight integration without special procedures means I don't 
need to determine "Start proc Args" and "Stop Proc Args" in the Parallel 
Environment Configuration. I just did gave them the value "none" and now 
the jobs are executed. Thanks.

But I have another problem: the mpi communication should go over the 
infiniband-network and not over the management network (at the moment it 
does the last one). To do so should I change the names in the /etc/hosts 
file (the hostnames and the management interfaces have the same names, 
and SGE selects hostnames not interfaces)? Or is there another possibility?

Ursula