Parallel running and runnning errors

Hi,

We do not support HPC type of cluster. We actually need someone with access to this kind of cluster to make the implementation.

Cheers,

Olivier
On 02 Jun 2015, at 02:07, Huanian Zhang <email address hidden> wrote:

> New question #267654 on MadGraph5_aMC@NLO:
> https://answers.launchpad.net/mg5amcnlo/+question/267654
>
> Hi Professor,
>
> Recently I am working on university's HPC account (High Performance Computer), but each time when I run the MG5, it has some errors like below:
>
> Command "generate_events " interrupted with error:
> error : can't start new thread
> Please report this bug on https://bugs.launchpad.net/madgraph5
> More information is found in '/gsfs1/xdisk/fantasyzhn/MG5_aMC_v2_2_2/ttz/run_03_tag_1_debug.log'.
> Please attach this file to your report.
>
> The compiler on machine is gcc (gfortran) 4.4.4, python 2.7.3, how do I solve this problem?
>
> I have another question about the parallel running. I need to generate a huge number of events. For instance, I need to generate 100 million ttbar events, if I use the multi_run command, it will take a few weeks and also it can not make the most use of the cpu on the node. I can request tens of Gb cpu memory on one node, but the serial running only use less than 1 Gb. How do I run the MG to make the most of cpu to speed it up?
>
> Thank you very much.
>
> Huanian Zhang
>
>
>
> --
> You received this question notification because you are an answer
> contact for MadGraph5_aMC@NLO.

Hi Olivier,

Thanks a lot. But how do I speed it up in my local computer? Since I need to run enormous events, the speed means a lot to me.

My local computer has 8 processors, each process has 4 cores, total cpu memory is 32 Gb, it is Intel Core i7-3770 @ 3.4GHz.

Thank you very much.
Huanian

Hi,

The default running mode is multi-core with the maximum number of core available.
If the code fails to find that number of core. You can force it by editing the file
input/mg5_configuration.txt

and change the line
# nb_core = None
to
nb_core = 12

also checks that you have either the line defining “run_mode” is either commented or on “2“

Cheers,

Olivier

On 02 Jun 2015, at 22:16, Huanian Zhang <email address hidden> wrote:

> Question #267654 on MadGraph5_aMC@NLO changed:
> https://answers.launchpad.net/mg5amcnlo/+question/267654
>
> Status: Answered => Open
>
> Huanian Zhang is still having a problem:
> Hi Olivier,
>
> Thanks a lot. But how do I speed it up in my local computer? Since I
> need to run enormous events, the speed means a lot to me.
>
> My local computer has 8 processors, each process has 4 cores, total cpu
> memory is 32 Gb, it is Intel Core i7-3770 @ 3.4GHz.
>
> Thank you very much.
> Huanian
>
> --
> You received this question notification because you are an answer
> contact for MadGraph5_aMC@NLO.

Thanks Olivier Mattelaer, that solved my question.

Hi,

I have access to a HPC cluster in my Institution. In fact, I'm interested in this possibility since I need to generate a large number of events and it would be perfect if I can use the HPC cluster of my institution.

Bests,
Juan

Hi Juan,

It depends of the type of job scheduler that you have in your HPC cluster.
we support PBS, SGE, condor, LSF, …

If yours is not in the list, it might be possible to implement it following the instructions:
https://answers.launchpad.net/mg5amcnlo/+faq/2249
Note if you have an infrastructure type BlueGen then this might be another business

Cheers,

Olivier

> On Jan 27, 2016, at 11:13, Juan Carlos Vasquez Carmona <email address hidden> wrote:
>
> Question #267654 on MadGraph5_aMC@NLO changed:
> https://answers.launchpad.net/mg5amcnlo/+question/267654
>
> Juan Carlos Vasquez Carmona posted a new comment:
> Hi,
>
> I have access to a HPC cluster in my Institution. In fact, I'm
> interested in this possibility since I need to generate a large number
> of events and it would be perfect if I can use the HPC cluster of my
> institution.
>
> Bests,
> Juan
>
> --
> You received this question notification because you are an answer
> contact for MadGraph5_aMC@NLO.