Data Consistency for Self-acting Load Balancing of Parallel File System

With the increasing scale of I/O servers, nice load balancing algorithm among I/O servers is becoming more and more important in parallel file systems. However, seldom systems have focused their attentions on the imbalance between I/O servers. In order to address this problem, we have proposed a self-acting load balancing strategy. In this paper, data consistency between I/O servers is mainly discussed. A method called re-migration is adopted to reduce the concurrent data-access impact on the self-acting load balancing strategy. According to some experiments, this method is proved to be able to reduce the average response time and improve performance of parallel I/O obviously. Besides, it also improves the flexibility of systems.