Speaker
Andrew Haas
(SLAC National Accelerator Laboratory)
Description
We are now in a regime where we observe substantial multiple proton-proton collisions within each filled LHC bunch-crossing and also multiple filled bunch-crossings within the sensitive time window of the ATLAS detector. This will increase with increased luminosity in the near future.
Including these effects in Monte Carlo simulation poses significant computing challenges. We present a description of the standard approach used by the ATLAS experiment and details of how we manage the conflicting demands of keeping the background dataset size as small as possible while minimizing the effect of background event re-use. We also present details of the methods used to minimize the memory footprint of these digitization jobs, to keep them within the grid limit, despite combining the information from thousands of simulated events at once.
We also describe an alternative approach, known as Overlay. Here, the actual detector conditions are sampled from raw data using a special zero-bias trigger, and the simulated physics events are overlaid on top of this zero-bias data. This gives a realistic simulation of the detector response to physics events. The overlay simulation runs in time linear in the number of events and consumes memory proportional to the size of a single event, with small overhead. We explain the computational issues and challenges that will arise in running overlay in production mode on the grid. Finally we discuss the computational issues that may arise in the future in generating large amount of luminosity weighted zero-bias data and making it available on the grid
Primary author
Collaboration Atlas
(Atlas)
Co-authors
Daniel Froidevaux
(CERN)
Philip Clark
(University of Edinburgh (GB))
