Research and Exploration of Mechanism for Reconfigurable Multi-Cluster System

Multi-cluster system has been a popular platform for HPC (high performance computing) environments. But it is lack of flexibility and usability, especially for the vehicle-bone and mobile parallel computation. To solve the problem, we propose a reconfigurable mechanism for the management of multi-cluster system in this paper. The concept of reconfiguration is that the structure of the multi-cluster system could change dynamically, including addition and quit of clusters. According to this, we analyze the key issues and design the reconfigurable mechanism by two aspects, including the architecture and the cases of the mechanism. Then some policies are proposed to achieve it, which are accomplished by aggregation algorithm and optimized Bully algorithm. This mechanism increases the flexibility of HPC environments.