Even though grid research is prosperous in an extensive context, few grid platforms for high performance computing are practical and in operation so far. Since most applications are moderate ones, trying to decompose these applications among distributed supercomputers will results in complexity in programming and optimizing, and heavy cost of communications. In this paper we advocate computing pool which shares distributed supercomputers but does not decompose applications. Analysis proves that it is a practicable computing platform, which can greatly improve quality of service and utilization of resources.