multicore in cluster mode of run

Asked by Juhi Dutta

Hi,

I have a query regarding running mg5 in the cluster mode.After generating the process and storing it in the output folder , I edit me5_configuration.txt file . one of the changes include :

#!Default Running mode
#! 0 : single machine/ 1: cluster /2:multicore
#run_mode= 1

In this case am I allowed to use multiple cores to give runs in the cluster? or only mode : 2 (which is the default one) allows usage of multicores for faster processing? I am not sure if I am correct in assuming that we should not use multicores in the cluster mode since in that case large runs composed of multiple short runs will take a longer time.

Kindly help me clear my understanding in this regard.

Thanks & Regards,
Juhi

Question information

Language:
English Edit question
Status:
Solved
For:
MadGraph5_aMC@NLO Edit question
Assignee:
No assignee Edit question
Solved by:
Olivier Mattelaer
Solved:
Last query:
Last reply:
Revision history for this message
Best Olivier Mattelaer (olivier-mattelaer) said :
#1

Dear Juhi,

> #!Default Running mode
> #! 0 : single machine/ 1: cluster /2:multicore
> #run_mode= 1

You need to uncomment the line in order to have that line to apply.
The default is that all line are commented (=> use the default value).

Madgraph splits the computation of the integrals in various channel of integration.
Each of those are completely independent of the other one, and can therefore be run independently.
(Either on a different core or on a different machine).

The three mode of submission, correspond to various way to handle this list of jobs.
In the first case (=0) we just run them locally one after the other.
In the second case (=1) we just send them all to the cluster scheduler and wait them to finish (and then decide if we need to re-submit some job or not in order to reach the target)
The third case (=2), we just ask the computer to run N jobs, and as soon as one id finished, we ask the system to run another one.

So in cluster mode, we do not have need of multicore machine since in our case 1job = 1 unit of computation.
If your cluster is mainly designed to have 1 job running on a multicore machine then you need to read the following page, in order
to have madgraph running in a more efficient way:
https://answers.launchpad.net/mg5amcnlo/+faq/2249

My suggestion would be the add a buffer and then pack N madgraph jobs as a single one, which will just run them in parallel on the multicore machine. Maybe not 100% efficient but that should already help quite a lot.

Cheers,

Olivier

On 12 Apr 2015, at 17:06, Juhi Dutta <email address hidden> wrote:

> New question #265070 on MadGraph5_aMC@NLO:
> https://answers.launchpad.net/mg5amcnlo/+question/265070
>
> Hi,
>
> I have a query regarding running mg5 in the cluster mode.After generating the process and storing it in the output folder , I edit me5_configuration.txt file . one of the changes include :
>
> #!Default Running mode
> #! 0 : single machine/ 1: cluster /2:multicore
> #run_mode= 1
>
> In this case am I allowed to use multiple cores to give runs in the cluster? or only mode : 2 (which is the default one) allows usage of multicores for faster processing? I am not sure if I am correct in assuming that we should not use multicores in the cluster mode since in that case large runs composed of multiple short runs will take a longer time.
>
> Kindly help me clear my understanding in this regard.
>
> Thanks & Regards,
> Juhi
>
>
>
> --
> You received this question notification because you are an answer
> contact for MadGraph5_aMC@NLO.

Revision history for this message
Juhi Dutta (dutta-juhi91) said :
#2

Hi,

Thank you for pointing out about the #.I missed it.

Also regarding the suggestion,I have just started fiddling with the cluster mode and am quite unfamiliar with a lot of settings.
I will hopefully try it in some time.I will revert to this question if I encounter any problems.

Thanks & Regards,
Juhi

Revision history for this message
Juhi Dutta (dutta-juhi91) said :
#3

Thanks Olivier Mattelaer, that solved my question.