Research On Journal Based High-availability Mechanism Of Cluster File Systems

Now,the cluster system has matured as a mainstream method for building high performance computers. A lot of cluster file systems spring out in recent years. Cluster file system provides high-performance, scalable and high-availability file service for applications. Because the complication of the cluster file system, the high-availability technology would be a key feature for a successful file system. In this paper we will introduce the journaling mechanism which can be used by the cluster file system to provide high-availability file service with little performance lose. The crucial contributions are as follows:1. In this paper we list the requirements and techniques on high-availability of different cluster file systems, then the journaling mechanism introduced in this paper, including the key issues and the strategies, also prove the correctness of this mechanism.2. We used this mechanism in the design of the second version of Dawning Cluster File System(DCFS2) which developed by NCIC for a 64-bit Linux cluster called Dawning 4000A can provide high performance, large file system support and easy to management, and also the high-availability. In this paper, we introduce the design and implementation of the high-availability of DCFS2, including the high-availability of the metadata servers and the high-availability of the DCFS2 client-side file system.3. We use levels to define the high-availability of a system. Because we cannot do the quantitative analysis for the degree of availability in software system like hardware system, this dissertation does some researches and analysis on the cluster file system's high availability issue from a point of qualitative view. Based on the apply mode of distributed file system, we did some clustering analysis for the factors which will influence the high availability issue. Based on the failure and restoring factors, we defined the high availability in different levels.4. Performance experiments on DCFS2 show that DCFS2 can provide competitive bandwidth and metadata performance and high-availability. The time spent during the recovery is also tested.The thesis also discusses some future work like the support of failure recovery when multiple parts in the file system crash and so on.