MadGraph5_aMC@NLO

Gridpacks multiple instances

Asked by Amin Aboubrahim

Hi Olivier,

1) I am using gridpacks to generate tt~ samples. The thing is that I cannot run separate jobs of the same gridpack on several nodes. It seems that madevent cannot have multiple instances.
Is it possible to override this and get rid of RunWeb? It seems odd to use the gridpack only once per node.

2) Also I am generating some QCD samples and are divided in bins of HT. I have noticed that higher bins require more time to generate the same number of events. Does that make sense? My first bin was quick but the others are taking too much time.

Thank you,
Amin

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
Olivier Mattelaer (olivier-mattelaer) said : 		#1

Hi Amin,

1) It is possible in 2.6.1 if you set the full gridpack directory in read-only mode ( you can use a chmod for that).
In that case, you have to execute the gridpack from different (fully empty) directory and call the executable from that directory.
The generated events will then be created in each directory.

2) This is a typical phase-space integration problem, by adding cut you increase the variance of the function and therefore you need to evaluate more matrix-element to reach the same level of accuracy. Did you use the latest version of the code? In that one, we try to include information from such cuts in a smarter way to speed-up the generation but this is not even a factor of two speed up.

My advise if you need large HT bin would be to investigate the biassing option.

Cheers,

Olivier

Revision history for this message
Amin Aboubrahim (amin83) said : 		#2

Hi Olivier,

1) I forgot about that. I have seen it implemented before and with older versions too. From what you're saying it looks like it is only implemented in 2.6.1 ??

2) I am currently using 2.6.0. I'm not familiar with biasing. I need to read about this.

3) One more question: when the gridpack generation is over, the summary that appears on the html page seems to include a number under "Events". Now the command line screen shows zero events generated but the html shows something else. What is that number?

Thanks a lot.

Best,
Amin

Revision history for this message
Best Olivier Mattelaer (olivier-mattelaer) said : 		#3

Hi Amin,

1)I confirm that a gridpack cannot be in read-only mode before 2.6.1. This is a fully new option that cannot be working before (this is unlikely that some non-core developers created a patch for this since this was related to deep change in the code).

3) The number of events set on the command line screens is hardcoded to zero to avoid confusion.
But in the code computing cross-section and generating events are correlated (you can not really do one without the other one).
So for each channel, we have some events generated and the associated efficiency to generate those and those number appears in the html indeed.

Revision history for this message
Amin Aboubrahim (amin83) said : 		#4

Thanks Olivier Mattelaer, that solved my question.

Revision history for this message
Wenxiao Wang (wenxiao) said : 		#5

Dear Oliver,

Sorry to resurrect this thread but I'm having problems with the first question which is about running parellel jobs with the same gridpack.
###################################
1) It is possible in 2.6.1 if you set the full gridpack directory in read-only mode ( you can use a chmod for that).
In that case, you have to execute the gridpack from different (fully empty) directory and call the executable from that directory.
The generated events will then be created in each directory.
###################################
==>I'm a newbie to MadGraph and apologise if my question is dumb. Could you please give me more information about this feature and how exactly I should make it work? I tried to change the "madevent" folder(which is extracted from the gridpack package) to readonly by "chmod -R 555 madevent", then executed the script "run.sh" from an empty folder, but it crashed with IOError like saying "No such file or directory: xxxxxxxxxx/results.dat". I'll be very appreicated if you can help me with this issue.
Thanks a lot!

Best regards,
Wenxiao

Revision history for this message
Olivier Mattelaer (olivier-mattelaer) said : 		#6

Hi,

Looks like you did everything as I was expecting.
Do you have any log file in your xxxxxxxx directory?
it might indicate why the results.dat is not created correctly.

to be more detailed to make my tests I did:

> cd ~madgraph/2.6.1
> ./bin/mg5
> genrate p p > t t~
> output GP1
> launch GP1
> set gridpack T
>
> cd /tmp
> mkdir RO
> mkdir RO_test
> cd RO
> cp ~madgraph/2.6.1/GP1/run_01_gridpack.tar.gz .
> tar -xzpvf run_01_gridpack.tar.gz
> chmod -R 555 madevent
> ../RO_test
> ../RO/run.sh 100 200
> ls
> #-> events.lhe.gz

Maybe you can try the same suite to see if tis one works in your case.

Cheers,

Olivier

> On 20 Mar 2018, at 02:28, Wenxiao Wang <email address hidden> wrote:
>
> Question #662721 on MadGraph5_aMC@NLO changed:
> https://answers.launchpad.net/mg5amcnlo/+question/662721
>
> Wenxiao Wang posted a new comment:
> Dear Oliver,
>
> Sorry to resurrect this thread but I'm having problems with the first question which is about running parellel jobs with the same gridpack.
> ###################################
> 1) It is possible in 2.6.1 if you set the full gridpack directory in read-only mode ( you can use a chmod for that).
> In that case, you have to execute the gridpack from different (fully empty) directory and call the executable from that directory.
> The generated events will then be created in each directory.
> ###################################
> ==>I'm a newbie to MadGraph and apologise if my question is dumb. Could you please give me more information about this feature and how exactly I should make it work? I tried to change the "madevent" folder(which is extracted from the gridpack package) to readonly by "chmod -R 555 madevent", then executed the script "run.sh" from an empty folder, but it crashed with IOError like saying "No such file or directory: xxxxxxxxxx/results.dat". I'll be very appreicated if you can help me with this issue.
> Thanks a lot!
>
> Best regards,
> Wenxiao
>
> --
> You received this question notification because you are an answer
> contact for MadGraph5_aMC@NLO.

Revision history for this message
Wenxiao Wang (wenxiao) said : 		#7

Dear Oliver,

Thanks for your reply.
I tried to follow exactly your test and found that it works well. However, if I change the number of events I'd like to generate, say "../RO/run.sh 1 200", the error appears again: IOError: [Errno 2] No such file or directory: "P1_gg_ttX/G1/results.dat". According to my test, the error would be produced with nevent = 1 up to 6. Could you please comment on this? Thanks.

Best regards,
Wenxiao

Revision history for this message
Wenxiao Wang (wenxiao) said : 		#8

Dear Oliver,

Maybe you missed my last post, but could you please comment on that? Thanks a lot!

Best regards,
Wenxiao

Revision history for this message
Olivier Mattelaer (olivier-mattelaer) said : 		#9

Hi,

yes sorry:

here is a patch for this:

=== modified file 'madgraph/interface/madevent_interface.py'
--- madgraph/interface/madevent_interface.py 2018-02-22 13:49:53 +0000
+++ madgraph/interface/madevent_interface.py 2018-03-27 19:37:09 +0000
@@ -6456,6 +6456,7 @@
         #print 'run combine!!!'
         #combine_runs.CombineRuns(self.me_dir)

+ return
         #update html output
         Presults = sum_html.collect_result(self)
         cross, error = Presults.xsec, Presults.xerru

Now just for the fun, I have timed the gridpack to generate N events
N=1 -> 0.8s
N=10 -> 2.5s
N=100 -> 2.5s
N=1000 -> 3.32s
N=10k. -> 27.1s
N=100k -> 6m25s

Cheers,

Olivier

> On 26 Mar 2018, at 11:03, Wenxiao Wang <email address hidden> wrote:
>
> Question #662721 on MadGraph5_aMC@NLO changed:
> https://answers.launchpad.net/mg5amcnlo/+question/662721
>
> Wenxiao Wang posted a new comment:
> Dear Oliver,
>
> Maybe you missed my last post, but could you please comment on that?
> Thanks a lot!
>
> Best regards,
> Wenxiao
>
> --
> You received this question notification because you are an answer
> contact for MadGraph5_aMC@NLO.