Hi,
you requested 4 in JDL nodes, but you got 16 processes in the same host.
How many cpus/cores per node do you have? Is PBS configured correctly
to see this number of cpus?
Can you show the a qstat -f of the job so we can check how they were allocated?
Regards,
Enol.
On Tue, Nov 23, 2010 at 5:09 PM, Claudiu Demian
<[log in to unmask]> wrote:
> Hi,
>
> These are the files:
> http://ui01.mosigrid.utcluj.ro/~demi/demi_UhAD3uYapjCcgAJ7M44kBg/mpi-start.err
> http://ui01.mosigrid.utcluj.ro/~demi/demi_UhAD3uYapjCcgAJ7M44kBg/mpi-start.out
>
> Still, the same.
>
> Claudiu
>
>
> On 11/23/2010 05:53 PM, Nikolay Kutovskiy wrote:
>> Claudiu Demian wrote on 23/11/10 18:45:
>>> Now the jobs sent are successful. When I submit them directly with
>>> glite-ce-job-submit the job is sent to all 4 nodes (verified that by
>>> inserting sleep(60) in hello.c) but the output still shows output from
>>> one node only).
>>> when I submit them with glite-wms-job-submit the job is also successful
>>> and spread to 4 nodes (according to glite-wms-job-status ) but I can't
>>> figure a way to get the output files back.
>> In case the job was submitted via WMS use
>> $ glite-wms-job-output --dir <path_to_store_the_output> <jobid>
>>
>> Using that command you get all files mentioned in OutputSandbox
>> attribute of your jdl file.
>>
>> My test jdl file looks like below:
>> $ cat tests/mpi_tests/hello/hello-openmpi.jdl
>> JobType = "Normal";
>> CPUNumber = 3;
>> Executable = "mpi-start-wrapper.sh";
>> Arguments = "hello OPENMPI";
>> StdOutput = "hello-openmpi.out";
>> StdError = "hello-openmpi.err";
>> InputSandbox = {"mpi-start-wrapper.sh","mpi-hooks.sh","hello.c"};
>> OutputSandbox = {"hello-openmpi.err","hello-openmpi.out"};
>>
>> HTH,
>> Nikolay.
>>>
>>> Any other suggestions as to what I am doing wrong?
>>>
>>> Thanks,
>>> Claudiu
>>
>
|