Efficiency of resource brokering in grids for high energy physics computing

This paper presents a study of the efficiency of resource brokering in a computational grid constructed for CPU and data intensive scientific analysis. Real data is extracted from the logging records of an in-use resource broker relating to the running of Monte Carlo simulation jobs, and compared with detailed modeling of job processing in a grid system. This analysis uses performance indicators relating to how efficiently the jobs are run, as well as how effectively the available computational resources are being utilized. In the case of a heavily loaded grid, the delays incurred at different stages of brokering and scheduling are studied, in order to determine where the bottlenecks appear in this process. The performance of different grid setups is tested, for instance homogeneous and heterogeneous resource distribution, and varying numbers of resource brokers. The importance of the speed of the grid information services is also investigated.