| The challenge to enterprise massive storage system in the network is aggrevating: the explosive increase of digitalization information, the increase of data significance and security, the tremendous pressure of storage performance to the great number of users and multimedia applications, higher availability and maintainability of 24×7 service are all putting increasing pressure on enterprise massive storage system. The major techniques such as the high performance storage device (RAID), Network Attached Storage (NAS) and Storage Area Network (SAN) have its merit and deficiency respectively. The cause of this problem is that physical and logical organization structure of storage system is a static organization, which can't adapte to continual change outside workload. The organization structure of storage system always is fit for given applications, but failures to adapt to variational mechanism of storage requirement.First of all, Based on the concept and key idea of evolving storage system, the evolving strategy is made of physical evolution and logistic evolution. By detailed discussing main methods of implementing physical evolution and logistic evolution and researching hardware architecture and software structure, we build important foundation for design and implement evolving storage system.ESS needs to make its organization structure and kinds of attribute data distribution change according to in-out data stream and to enhance its performance. On condition that it dynamically find and predict the storage areas requested frequently. So, we not only introduce a new sequence degree-based clustering algorithm to find the frequently accessed storage areas, but also adopt ARMA time series model to forecast the storage areas requested frequently by future I/O requests. To address the problem of accurate forecast, this paper adopts dynamic parameter estimation policy to ARMA model. This method has very higher matching ratio and can sufficiently guide the adjustment of storage system. At the same time, based on characteristic of ESS data distribution and I/O request area, we can improve storage system performance by prefetching all data of read request at the next time in idle time of storage system. Now some researches on optimizing data organization structure and distribution to storage system are basically that data in hotspot area are move to some areas that have higher performance. This kind of strategy will lead data mapping to very complexity, and make data mapping become performance bottleneck. So we present a method that can transform data organization structure and distribution according to workload characteristic, at the same time, it can reduce data moving. In order to simplify data mapping complexity and mitigate server workload, we adopt the two-class mapping mechanism to improve mapping efficiency based on ESS characteristic.We presents a performance threshold-based optimize strategy according as data organize structure and distribution of ESS. By analyzing action of RAID1 and RAID5 under synchronous workload and Closed Forked-Join Queue Network model (CFJQN), we deducing a approximation analytical expression to predict disk array mean response time and compute threshold of parameters for RAID organization transformation optimize stripe unit size transformation.Based on the system architecture characteristic of evolving storage system, in order to ensure data availability and data migration performance at evolving storage system, this paper recommend a method which can implement snapshot and serverless backup by inserting a virtual device layer at server and storage node. It made storage nodes have"no window"backup and high performance data migration, improved storage system availability and backup performance.
|
PDF Full Text Request