Self-acting load balancing with parallel sub file migration for parallel file system

With number of I/O servers increasing, the load imbalance among them becomes an emerging challenge in using highend computing system effectively. But most of contemporary file systems like PVFS2 and Lustre lack load balancing to optimize parallel I/O performance when facing large number of I/O servers. To address this problem, this paper proposes self-acting load balancing for parallel file systems. The proposed self-acting load balancing builds on dynamic file migration, and borrows the ideas from intelligent computing to enable automatic load balancing. Furthermore, in order to support promising scalability, the self-acting load balancing is fundamentally based on distributed architecture, which also permits parallel file migrations to accelerate balancing progress. By taking PVFS2 as basic platform, we implement the self-acting load balancing, and the experiments show that PVFS2 with self-acting load balancing can reduce the average response time and improve performance of parallel I/O obviously compared with original PVFS2.