LHCb: Statistical Comparison of CPU performance for LHCb applications on the Grid

The usage of CPU resources by LHCb on the Grid id dominated by two different applications: Gauss and Brunel. Gauss the application doing the Monte Carlo simulation of proton-proton collisions. Brunel is the application responsible for the reconstruction of the signals recorded by the detector converting them into objects that can be used for later physics analysis of the data (tracks, clusters,…) Both applications are based on the Gaudi and LHCb software frameworks. Gauss uses Pythia and Geant as underlying libraries for the simulation of the collision and the later passage of the generated particles through the LHCb detector. While Brunel makes use of LHCb specific code to process the data from each sub-detector. Both applications are CPU bound. Large Monte Carlo productions or data reconstructions running on the Grid are an ideal benchmark to compare the performance of the different CPU models for each case. Since the processed events are only statistically comparable, only statistical comparison of the achieved performance can be obtained. This contribution will present the result of such comparison from recent LHCb activities on the Grid. The result are compared for different CPU models and the dependence with the CPU clock is shown for CPUs of the same family. Further comparisons with HEPIX WG results and LHCb, and other LHC experiments, benchmarking are also included.