MadGraph for the Grid

Scattered information

Here is a copy and paste of first Tim's email on the MadGraph4Grid. Outdated! Needs revision...

Hi

I have made progress on creating scripts for "grid" running. Below is a printout of the file in
madgraph@madgraph.uiuc.edu/MG_ME_Grid/GridReadme.txt.

Until everything is fully integrated, I just created a script to "gridify" a process. That makes it relatively easy to play/test w/o making any changes to our templates etc.

Basically you can generate a process on the web,
then from a terminal "gridify it", then when you run from the web it will do a "warm-up". Once that is done you can generate "jobs" that are stand-alone and generate unweighted events that can be combined together.  I'll work on optimizing and testing more. But wanted you to know the status in case anyone wants to start playing with it.

Tim


Creating a grid-ready process

******* Initialization (done once) *********

1) Generate process as normal (via web)

2) Open terminal window and go to base directory for process
  (e.g. MadGraphData/tstelzer@uiuc.edu/PROC30 )

3) run  ~madgraph/bin/gridify
(Note this updates several files... but all are compatable
with the standard local runs except that it comments out the refine
call from bin/generate_events. )


4) run generate events (via web)

You now have a "good" run and plots etc. This will be used for
determining the # of events to collect from the different channels.


******* Event Generation (repeat as many times as necessary) *****

1) In terminal window run  ./bin/gen_grid_jobs
  (this will create gridpack.tar.gz)

2) unpack gridpack.tar.gz on grid computer

3) run bin/run_grid_jobs   on grid computer

4) Specified number of unweighted events will be in SubProcesses/unweighted_events.lhe


************************* Parameters ************************

Cards/grid_card.dat
     GridRun = .true.
This tells symmetry.f, gen_ximprove.f and combine.f if this
is a standard run, or a grid run. (.false. will result in standard run)

Source/run_config.inc
     double precision acc_wu
     parameter       (acc_wu = 0.01)   !Desired accuracy from warmup run
     integer      npoints_wu,        itmax_wu !warmup # points/iterations
     parameter   (npoints_wu = 4000, itmax_wu = 8)
     integer      min_gevents_wu
     parameter   (min_gevents_wu=200)   !Minumum # unweighted events to generate from chan=

-- Main.FabioMaltoni - 07 Nov 2007