Hi Enol,
On 11/26/2010 04:52 PM, Enol Fernández del Castillo wrote:
> Hi Claudiu,
>
> On 11/26/2010 03:12 PM, Claudiu Demian wrote:
>> Hi Enol,
>>
>> You were right, I had that line.
>> I've changed it according to your indications. This is my output now:
>>
>> # echo "#PBS -l nodes=64" | perl /var/spool/pbs/submit_filter.pl
>> #PBS -l nodes=1:ppn=16+48:ppn=1
>
> Still does not look right, that's one node with 16 cpus and 48 with 1,
> but in a previous mail you were saying that you have only 4 nodes with
> 16 cpus each. So the expression should be like nodes=4:ppn=16.
At our cluster, we have 64 nodes with 16 CPUs each. I only wanted to use
4 of the nodes with all of their 64 CPUs. I am sorry for the confusion.
I guess the output should be nodes=64:ppn=16. (nodes=63 actually, since
one of them is in service).
>
>>
>> If I try to run a job now, it exists with a compiling error:
>>
>> -<START PRE-RUN HOOK>---------------------------------------------------
>> Compiling /home/ops053/home_cream_266664921/CREAM266664921/hello
>> mpicc -o /home/ops053/home_cream_266664921/CREAM266664921/hello
>> /home/ops053/home_cream_266664921/CREAM266664921/hello.c
>> Error compiling program. Exiting...
>
> Have you installed mpicc in all the WN? Which error gives the compilation?
Yes, mpicc is installed on all of them. All the nodes are identical,
since we use a provisioning system based on puppet. The output of the
job is in the following directory:
http://ui01.mosigrid.utcluj.ro/~demi/demi__eje7VNQJnWbhtJpZdPPvQ/
Claudiu
>
> Enol.
>
>>
>> Claudiu
>
|