Concepts for fast large scale Monte Carlo production for the ATLAS experiment

The huge success of Run 1 of the LHC would not have been possible without detailed detector simulation of the experiments. The outstanding performance of the accelerator with a delivered integrated luminosity of 25 fb-1 has created an unprecedented demand for large simulated event samples. This has stretched the possibilities of the experiments due to the constraint of their computing infrastructure and available resources. Modern, concurrent computing techniques optimized for new processor hardware are being exploit to boost future computing resources, but even most optimistic scenarios predict that additional action needs to be taken to guarantee sufficient Monte Carlo production statistics for high quality physics results during Run 2. In recent years, the ATLAS collaboration has put dedicated effort in the development of a new Integrated Simulation Framework (ISF) that allows running full and fast simulation approaches in parallel and even within one event. We present the main concepts of the ISF, which allows a fine-tuned detector simulation targeted at specific physics cases with a decrease in CPU time per event by orders of magnitudes. Additionally, we will discuss the implications of a customized simulation in terms of validity and accuracy and will present new concepts in digitization and reconstruction to achieve a fast Monte Carlo chain with a per event execution time of a few seconds