Faq t2b: Difference between revisions

From T2B Wiki
Jump to navigation Jump to search
Line 30: Line 30:


=== MadGraph taking all the cores of a workernode ===
=== MadGraph taking all the cores of a workernode ===
The default settings for MadGraph is to take all the available cores. This kills the site.  
The default settings for MadGraph is to take all the available cores. This kills the site.
 
that is why you need to uncomment and set 2 variables in the '''md5_configuration.txt''' file (not the '''dat''' file), '''run_mode''' & '''nb_core'''.<br>
The run mode should be set to 0, single machine, via:
run_mode = 0
 
If the number of cores used by MadGraph is higher than 1, this needs to be asked to the job scheduler with the following directive added to your HTCondor submit file:
If the number of cores used by MadGraph is higher than 1, this needs to be asked to the job scheduler with the following directive added to your HTCondor submit file:
<pre> request_cpus = "2" </pre>
<pre> request_cpus = "2" </pre>
Where ppn is the number of cores you request. <br>
 
To tell MadGraph the number of cores he can take per job, use the following recipe:
To tell MadGraph the number of cores he can take per job, use the following recipe:
<pre>
<pre>
Line 40: Line 45:
save options
save options
</pre>
</pre>
or in the '''md5_configuration.txt''':
nb_core = 1
Note 'nb_core' and 'request_cpus' must alway be the same value! <br>
Note 'nb_core' and 'request_cpus' must alway be the same value! <br>
Note also that if you ask for more than one core your time in the queue will probably be longer as the scheduler needs to find the correct amount of free slots on one single machine. We advise against putting this number higher than one unless you really need it for parallel jobs.
Note also that if you ask for more than one core your time in the queue will probably be longer as the scheduler needs to find the correct amount of free slots on one single machine. <br>
We advise against putting this number higher than one unless you really need it for parallel jobs.

Revision as of 11:18, 3 October 2022



List of the UIs / mX machines:

- mshort: m1, m2 , m3 => 20 minutes of CPU time per process
- mlong: m4, m5, m6 , m7 => 1 hour of CPU time per process

Keep ssh connection to UI open:

Add option ' -o ServerAliveInterval=100 ' to your ssh command


Debugging SSH connection to mX machines:

  1. Check permissions on ssh keys:
> ll $HOME/.ssh
-rw------- 1 rougny rougny    411 avr 29  2019 id_ed25519
-rw-r--r-- 1 rougny rougny    102 avr 29  2019 id_ed25519.pub
To have the correct permissions:
chmod 600 $HOME/.ssh/id_ed25519
chmod 644 $HOME/.ssh/id_ed25519.pub
2. If that does not fix it, send us the output of those commands via chat/email:
> ll $HOME/.ssh
> date && ssh -vvv MYUSERNAME@m2.iihe.ac.be     <-- it needs to be on a specific machines (no mshort/mlong) so that we can read the logs!


MadGraph taking all the cores of a workernode

The default settings for MadGraph is to take all the available cores. This kills the site.

that is why you need to uncomment and set 2 variables in the md5_configuration.txt file (not the dat file), run_mode & nb_core.
The run mode should be set to 0, single machine, via:

run_mode = 0

If the number of cores used by MadGraph is higher than 1, this needs to be asked to the job scheduler with the following directive added to your HTCondor submit file:

 request_cpus = "2" 

To tell MadGraph the number of cores he can take per job, use the following recipe:

./bin/mg5_aMC 
set nb_core 1  #or 2 or whatever you want
save options

or in the md5_configuration.txt:

nb_core = 1

Note 'nb_core' and 'request_cpus' must alway be the same value!
Note also that if you ask for more than one core your time in the queue will probably be longer as the scheduler needs to find the correct amount of free slots on one single machine.
We advise against putting this number higher than one unless you really need it for parallel jobs.