Re: [Condor-users] How to get the machine file in parallel jobs

[miaocb <miaocb@xxxxxxx>]



Hi Imre，

  Thanks for sharing your solution with me.  I use windows, and I found that

the environment variable %_CONDOR_JOB_AD% refer to a .job.ad file，

and some useful information are provided in that file including the machines allocated to the job.

I  wrote a C program to obtain the machine list from that file. 

---

Chunbao Miao

 

From: Imre Szeberenyi

Date: 2012-08-27 19:36

To: Condor-Users Mail List

Subject: Re: [Condor-users] How to get the machine file in parallel jobs

Hi Chunbao,

 

I had the same problem before.

I have not found proper scripts for submitting in openmpi environment.

I have found one in the share/doc/condor-7.8.1/etc/examples/ directory 

which

installs ssh daemons on the remote machines, but I cannot use it in SMP

environment.

 

Finally I found a solution, may be it helps for you.

  - I created a shell script which collects the host info from job 

status and

    creates a host file containing job IDs and slot numbers for starting 

mpirun.

  - I force the mpirun to use  condor_ssh_to_job. The only problem is the

    mpirun checks the format of the host file and if it starts with 

numbers it

    assumes these are IP addresses. So I added a constant string to the job

    IDs and a wrapper starts the condor_ssh_to_job, which removes the 

constant string.

 

I enclosed my scripts, I hope you can find it useful as well.

 

If your are using openmpi-1.4 change the last command of 

condor_openmpi.sh script to

 

exec $MPIRUN --prefix $MPI_HOME --mca plm_rsh_agent 

$_CONDOR_SSH_TO_JOB_WRAPPER \

               --hostfile $_CONDOR_PARALLEL_HOSTS_FILE $@

 

 

Best,

 

Imre

 

 

2012.08.26. 15:00 keltezéssel, miaocb@xxxxxxx írta:

> Hi All,

>      I successfully configured condor to run parallel jobs, but I can't figure out how to get a machine file that can be used by mpiexec or mpirun to start MPI jobs. Is there an environment variable that refers to the machine file?

>

> thanks

>

> Chunbao Miao

> _______________________________________________

> Condor-users mailing list

> To unsubscribe, send a message to condor-users-request@xxxxxxxxxxx with a

> subject: Unsubscribe

> You can also unsubscribe by visiting

> https://lists.cs.wisc.edu/mailman/listinfo/condor-users

>

> The archives can be found at:

> https://lists.cs.wisc.edu/archive/condor-users/

 

 

 

 

 

_______________________________________________

Condor-users mailing list

To unsubscribe, send a message to condor-users-request@xxxxxxxxxxx with a

subject: Unsubscribe

You can also unsubscribe by visiting

https://lists.cs.wisc.edu/mailman/listinfo/condor-users

 

The archives can be found at:

https://lists.cs.wisc.edu/archive/condor-users/