Research On Dynamic Replica Management Strategies In Global Distributed Storage System

With the rapid development of information technology, data storage has become an important realm in research of computer science. More and more large-scale scientific research, engineering applications and commercial computing field produce huge amount of data that can not be stored in a single storage system. Meanwhile the rapid increase of information resource challenges the reliability and the expansibility of storage system, The global distributed storage system plus (GDSS plus) is a system which integrates all the scattered and distributed storage resources into unified logical storage view, providing large volume, high performance and high availability storage service.The reliability and the expansibility of storage system, the improvement of storage system's load balance are essential parts of distributed storage system. The high delay of wide-area network restricts user's rapid access to the distributed data. The users of distributed storage system are decentralized, the quantity information they accessing may be petabytes, and in wide-area, distributed computing environments, so it is inefficient for many users to access one file at the same time. The replication management of distributed storage system dynamic creates file replicas between heterogeneous storage system, reduces the distance of file movement, optimizes the speed of multi-user data access, decreases the user access latency and bandwidth consume , improves system load balance and performance.The replication management system of GDSS plus proposed a model based on the virtual organization (VO) consist of storage agents and the user characteristics. The system includes three modules: agent monitor, replicas transporter and replicas manager etc. Replicas creation strategies can create file replicas according to the idea of VO and queuing theory. The detail process of replicas creation inside and outside the VO domain will be presented. A a-synchronism replicas update strategies based on the version will be adopted, it tracks files update status, uses the Master-slave pattern, and update the file according to the VO hierarchy, which guarantee the consistency and the speed of update. The combination of static selection and dynamic selection will be used as the replicas select strategies between the VO domains. And the idea of queuing theory will be refer by the replicas select strategies inside the VO domain.The experiments show that when employing the replicas management algorithm can reduce access latency, increase robustness, reliability and performance of distributed storage system.