Abstract: By using the algorithm that can dynamically select computation cost and running a benchmark program with moderate computation cost in different grid nodes, the obtained MFLOPS value of CPU can be used as a parameter to compare different CPUs’ computational ability directly so that users can make choices among various computing resources by means of it. Our experiments have proved that this method can reflect a CPUs’ computational ability stably and sensitively and that using it to compare CPUs’ computational ability is proper and efficient.

Key words: grid, CPUs’ computational ability, monitoring and disc overing service