Final Report of the ATLAS Detector Simulation Performance Assessment Group

In this document we discuss the output of the Detector Simulation Performance Assessment Group. We list a number of methods which consume significant CPU time and/or allocate significant memory as first targets for potential improvements which are uncorrelated with the physics content of the simulation. We suggest several approaches for reducing CPU time, by reducing the number of steps or tracks simulated. To maintain physics accuracy for the estimation of second-order quantities, e.g. resolution, the CPU reduction should ideally be achieved while maintaining the important steps and tracks which contribute to the leakage of energy from an absorbing element into an active element (or vice-versa), or between active elements. An investigation is attempting to identify the most productive areas for reducing the number of steps and tracks simulated beyond the existing use of production thresholds. We have identified a number of existing Geant capabilities which can be exploited in order to reduce the number of simulation steps and, proportionally, the CPU time. A number of these carry an expected low or moderate impact on the physics accuracy of typical setups, and are strong candidates for investigation in the near and medium future. Others carry the risk of greater impact on physics accuracy, and will require greater investigation. We have started to investigate the relationship for electrons between energy deposition, kinetic energy and step size, in order to identify whether a fraction of steps could be eliminated with negligible or low impact on the location or quantity of energy deposition. The initial finding is that opportunities could exist to eliminate about 8% of the steps, responsible for about 1% of energy deposition; there is need to study the impact of this change and to determine the best way of implementing it. Deeper investigations of the relation between these quantities is planned. We consider the allocation and use of memory as an important factor for further performance improvement and a target area for short and medium-term investigation. Note: The Abstract and Section 4 comprise the Executive Summary