| With the development of economy and science, the computational tasks become more and more in the fields of weather forecasting, high-power physics, biomedicine and physical geography. Their specialities are: lots of computational data, complicated algorithms, large-scale and most difficult computation. There is two solutions: one is constantly to design and manufacture some advanced supercomputers that had more computational power to meet the needs of complicate applications. But it will be confined by technologies and funds in the future. The other is to scatter these heavy tasts to all of existing computation resources. The distributed system is called as Cluster Computing System that can offer high computational power. We designed and implemented a cluster computing system based on Linux with existing hardware and software in the laboratory. The cluster system can meet the needs that we delve into some parallel and distributed computational works. We argued in this paper cluster architecture, system resource collection and management, load leveling algorithms and job schedule algorithms. The main contributions of this paper include: (1) Has discussed current domestic and international research works of cluster technology. (2) Has analyzed Linux 2.4 kernel, system call, function call and MPI parallel computation environment. (3) Has implemented a cluster computing system based on Linux and MPI. (4) This system not only can deal with serial computing jobs commited by users, but also deal with parallel computing jobs commited by users. After finished the computation, the cluster system will return the results to user. (5) Has designed and implemented a cluster middleware, including job submission module, resource management module, job schedule module and load leveling module. Moreover, we implement SSI (system sigle image). (6) Has discussed some robust designs of cluster. (7) Has designed and implemented some applications of cluster to inspect the computational abilities of cluster.
|
PDF Full Text Request