We describe the offline computing system of the Belle experiment, consisting of a computing farm with one thousand IA-32 CPUs. Up to now, the Belle experiment has accumulated more than 120 fb$^{-1}$ of data, which is the world largest $Bbar{B}$ sample at the $Upsilon(4S)$ energy. The data have to be processed with a single version of reconstruction software and calibration constants to perform precise measurements of $B$ meson decays. In addition, Monte Carlo samples three times larger than the real beam data are generated. To fullfill our computing needs, we have constructed the computing system with 90(300) quad(dual) CPU PC servers from multiple vendors as a central processing system. The details of this computing system and performance of data processing with the current model are presented.
Download