Research On Metadata-management Scheme For Object-based Storage Systems

In the object-based storage system, metadata server occupies an important position, because the capacity of metadata storage is small but the access is large and the management is very complex.In the existing load balancing strategy, the main way of selecting a server to store the copy of hot data is to calculate the server weight and to choose the server with the maximum weight. In this paper, we optimized this method with integrating the server's available bandwidth into the calculation of the weight values, so the molecular weight of the server includes server CPU, memory, hard drives, and available bandwidth. Furthermore, on the basis of server weight, we made the system choose the server according to the probability of the ratio of such server to all available servers,and it allows the servers to do their best. This strategy also works in the case of the user selecting the server to sending requests, and it could reduce the probability of congestion. In the end, we obtained the conclusion through theoretical analysis that, in the same conditions, using optimized approach proposed in this paper could shorten the users'average waiting time.Quick query of metadata information is the focus of metadata management. After explicit research into TBF (Two-level Bloom Filters), we proposed the architecture DBF (Double-level Bloom Filters) which could be used in the quick query of Multi-attribute metadata, and the approach GDA (Group-based Delayed-update Approach). In DBF, we add a memory cell after each attribute Bloom Filter to store attribute updating information temporarily, which could verify the attribute query and avoid error query caused by delaying attribute update. The main idea of GDA strategy is that: In time quantum T, the system does not execute the attribute updating operation immediately but store the update information in the memory cell temporarily. After time quantum T, the system will do the updating operation in each group contemporarily. GDA, in corporation with the verification device of DBF, will reduce the operation of attribution update greatly, not causing error query. Besides, we also put forward an"Edge-Numbers"Algorithm to solve the update conflicts among loops and tandems when the number of loops and tandems is larger than the number of shared edges, reduce the subtractive operation as much as possible and improve the update efficiency. The result of the simulation experiment shows that, Delayed-update Approach could significantly reduce the required operation of attribute updating.