| With the scale of storage resource's increasingly expanding, the diversity and multiplying of storage networks, the extensity and dynamical performance of resource distribution, all of these bring out new requirements and challenges in the management and accessing of data.The research on WAN-oriented storage resource discovery, aims at retrieval, discovery and matching on large-scale storage resource. Its purpose is to enhance resource using rate and service quality.Based on the practical application background of WAN-oriented storage resource discovery, this paper combines with resource retrieval methods in a peer-to-peer network and addresses a WAN-oriented storage resource discovery method, which includes its scalable architecture and techniques based on reinforcement learning.The architecture of storage resource discovery based on Storage Management Initiative Specification (SMI-S), adopts a peer-to-peer storage network model which expanded from service location protocol (SLP). Distributed hash table (Chord) is utilized to establish a structured peer-to-peer storage resource network, which is connected by efficient service communication between storage areas. In this way, the overhead of resource retrieval can be reduced and an efficient resource discovery can be carried out.History query and its feedback are used to build query-oriented routing indices step by step. Taking advantage of reinforcement learning, the storage resource discovery gradually approaches the real distribution of resources, and is able to accommodate itself to resource joining and quitting dynamically. Meanwhile, high scalability can be offered due to its low overhead of index maintenance and rapid learning process.A simulation of storage resource discovery process is built in this paper. The experimental result shows that, overhead can be reduced significantly while high recall is guaranteed. Also, our evaluation demonstrates that the system is suited for large-scale peer-to-peer storage resource network. |
PDF Full Text Request