Scientific Simulation Execution Support on a Closed Distributed Computer Environment

It is difficult for users to submit jobs to distributed computers and to retrieve calculation data from them in scientific computings. In this paper, we discuss and develop a robust job execution service system in a closed distributed computer system. The job execution service system consists of a dynamic system management servers, execution servers and data servers. The dynamic system management server is duplicated in order to keep the system robust, and has an assistant management server. The dynamic system management server has a function of the job execution system management, including software deployment, program compilation, job execution, job status retrieval and computing data retrieval. This system does not require special middleware such as Globus or UNICORE or GLite or so. Users access the web page on the dynamic system management server, and the clients submit jobs. After the submitted job finishes, the dynamic system management server collects the information from other distributed computers. The dynamic management server and its assistant server move dynamically to new servers, if the present servers become busy. The dynamic system management server also demands the execution server to transfer the result data to the optimal data server. The dynamic system management server copies the computing data and sends the compressed computing data to another optimal data server in order for a robust data storage system. The clients can deploy their programs, execute jobs and retrieve the result data by accessing only the web page in the job execution service system. This job execution management server also has a function of automatic system construction, so that the users can manage the setup of the job execution management system easily on their closed distributed computers.