Research On Fast Reconstruction, Scaling And Performance Optimization For Disk Array

With the coming of the era of information explosion, the disk array has gained rapid development as the primary information storage medium, and widely used in large scale storage systems. But there are some problems with itself that impact on performance and reliability. First, the disk failures can cause more frequent data reconstruction operations in disk array, while such reconstruction process may become very long when facing the increasing of online applications and single disk capacity. And thus developing fast reconstruction technology for disk array is imperative. Secondly, the capacity and bandwidth of disk array no longer meet the current storage trends with the fact of facing a rising data and requirements for storage performance. And How to rapidly scale the disk array becomes an urgent problem. And the third one is that with the development of cloud storage applications, disk array is increasingly being used in many various storage scenarios, especially sharing storage. Since disk array performance will decline as the front-end applications increase in sharing storage, therefore people have paid more attention to the disk array optimization technique to accommodate that scenario. Given that, we analyze and propose corresponding optimization methods for the disk array reconstruction, scaling and performance optimization, respectively.Due to the long serial reconstruction time of the current disk array, we propose a novel RAID data layout with parallel reconstruction feature, as called S2-RAID. It organizes disk partitions in parallel to create subRAID structures that can ensure reconstruction streams performing conflict-free I/O operations on each other, greatly reducing the reconstruction time. Meanwhile, S2-RAID can tolerate multiple disk failures under certain conditions, and also support parallel reconstruction in a lower granularity. The testing results show that S2-RAID speeds up data reconstruction time by a factor2-4compared to the traditional RAID layouts. At the same time the frontend performance can also be better guaranteed during the reconstruction.In the face of low reliability of the system due to long time of RAID scaling, we propose a fast RAID scaling method using a hybrid stripe layout, as called H-Scale. This method can significantly reduce data migration, disk I/Os, and parity updates during scaling process. The basic idea of H-Scale is to use the distribution characteristics of different stripe width, and re-mapping the old or new subRAID space to a uniform RAID space, with corresponding mapping algorithm for block addressing. Moreover, H-Scale can support multiple RAID levels and multiple RAID scaling, as well as many other features for an ideal RAID scaling. T The testing results for the H-Scale prototype show that, H-Scale outperforms traditional RAID scaling in terms of scaling time with an improvement of50%, and the frontend performance of H-Scale during scaling process is also improved by an average of20%. This method presents a good tradeoff between scaling time and frontend performance.Considering the wide application of sharing storage, we combed the actual problems of disk array in sharing storage and make a deep research based on the real world data sets, and present a RAID cache method-DROP to mitigate the disk array performance degradation. The principle of DROP is to build a cache buffer pool based on the free disk space using standard RAID level. The DROP system allocates this cache pool based on the access characteristics from different applications dynamically to balance the overall performance of the array, and the final evaluation results have achieved the expected goals.DROP allocates/manages a group of cache data areas and relocates/drops the portion of hot data at a predefined sub array that is a physical partition on the top of the entire shared array. By analyzing profiling data to make every cache area be monopolized by one server, we are able to determine optimal data relocation and partition of disks in the RAID to maximize large sequential block accesses on individual disks and at the same time maximize parallel accesses across disks in the arrayBy research on above three aspects, a more detailed analysis for the key issues effecting performance and reliability to the current disk array has been made and also some the targeted solutions are proposed, which has laid a good foundation for building disk array with the high performance and reliability.