A new formalism for dynamic reconfiguration of data servers in a cluster

摘要

The use of parallel file systems constitutes a high-performance solution to the problem known as I/O crisis in parallel or distributed environments. In the last years, clusters have become one of the most cheap and flexible frameworks for the deployment of parallel and distributed applications. Both parallel file systems and clusters have been successfully used in several scenarios, where it is possible to share and access data in an efficient way. In fact, clusters provide a huge number of advantages to this kind of systems, the wide availability of tools integrated with them being one of the most important. Nevertheless, clusters and, in general, high-availability distributed systems are characterized to be dynamically modified. Operations such as the addition or elimination of nodes are typical in a cluster environment. Therefore, it is necessary to use new approaches for the dynamic reconfiguration of the nodes that belong to a cluster. This paper describes a mathematical formalism for achieving high-performance and dynamic reconfiguration of data-based clusters with service maintenance.