MG takes toooooo long for just one channel

Asked by Yong Du on 2017-12-14

Hi Experts,

I have been trying to generate p p > w w w w in SM, but it seems it will never finish generating events-- it takes just too long, about 1h, to finish just one channel even with MadSpin (without MadSpin, the terminal looks like dead--no change at all at least for 1 hour).

Is there a way to speed up? I see an answer here (https://answers.launchpad.net/mg5amcnlo/+question/251103) saying that by setting the event number to be 100K, you can get a better timing. I set the event number to be 500K in the run card, and it seems there is no help at all... I only changed the event number in the default run card, everything else is untouched.

Any suggestion?

Thanks a lot,
Yong

Question information

Language:
English Edit question
Status:
Solved
For:
MadGraph5_aMC@NLO Edit question
Assignee:
No assignee Edit question
Solved by:
Olivier Mattelaer
Solved:
2018-07-17
Last query:
2018-07-17
Last reply:
2017-12-14

Dear Yong,

Well you have two choices here:

1) try to understand the reason why the code is slow (Is it a problem of too strong cut, or (integrable) singulartity that need to be cut, of ... This requires to look at the log of the channel which is quite slow and try to understand what is happening.
(first how the phase-space is generated, and then how the grid evolves)

2) Keep the code as it is but change the parralelization strategy to not be stuck with a single core running but split the code in many smaller piece to have a better use of your machine/cluster.
(in the appendix of the following paper:arXiv:1507.00020 , you will find the description of some of the hidden parameter of the run_card to increase the parralelizaton of the code

Cheers,

Olivier

Yong Du (fallive) said : #2

Thanks Olivier Mattelaer, that solved my question.