Distributed Lock Manager for Distributed File System in Shared-Disk Environment

摘要

Some new paradigms of large-scale distributed computing such as cluster, grid, and cloud computing have been recently developed to effectively support exponentially growing amount of data. Here numerous users store their data in the distributed storage that are accessed remotely anytime and anywhere. Therefore, an appropriate concurrency control such as locking is needed so that multiple users can collaborate with each other using the shared storage. This paper proposes a host lock mechanism based on distributed lock managers to support multiple clients in the clustered shared-disk environment. We have also introduced the object duplication scheme to minimize the waiting time for the request on exclusive mode, the reconstruction scheme for recovering the lock information after the lock server fails, and the deadlock avoidance scheme preventing the deadlock situation. Computer simulation reveals that the proposed mechanism is more flexible and efficient than the callback scheme typically employed for lock control and centralized locking mechanism. The communication cost of the proposed scheme is smaller than the callback scheme by adopting the blocking approach with a wait queue to eliminate spin-lock. The improvement gets more significant as the request rate of lock increases. It also significantly improves the processing time and utilization.