Research On APoRC Software Architecture Of RAID Controller

Now more and more businesses and individuals rely on the information stored in the computer system. The Mean Time To Failure(MTTF) value of current disk is getting higher and higher and the performance of the disk greatly improved, but the requirements of data storage is increasing at a tremendous speed and a enough large storage set is still subjected to frequent failures. The worse, to recovery a large storage system from the backup tape system need a few hours or even days or months, which makes the cost of storage system recovery very expensive.According to the existing technologies of disk array, the soft architecture based asynchronous recovery path (Asynchronous Path of Recovery Code, APoRC) mainly study the differences between the main code path and recovery path and provide a common soft architecture. The common soft architecture can dynamically adjust system performance according the load characteristics.The previous research has focused on the provision of allowing rapid prototyping and evaluation of architecture about redundant disk array, so far, the software architecture of RAID has not changed much and great deficiencies still exist on universal and dynamic performance adjustment. A new software architecture based on asynchronous recovery mechanism was proposed under the conditions of the commonality. The core of the current RAID was improved into two components, the first component is optimizer and the secondly component is implementation. A asynchronous recovery mechanism was used between the two components which can not be achieved in the traditional architecture and dynamically configure can be achieved in it, such can raise the flexibility of fault handling.The cache in RAID performance plays a very vital role. Parallel problems exist in read / write operations and some flaws exist in cache according to temporal locality and spatial locality. Addressing these problems, the past research can not focus on all of them, they can strengthen on the one hand, but relatively poorly on the other hand. The APoRC proposes a dual-cache mechanism which separate the read cache and write cache which allow users to read/write cache to execute parallelly, at the same time, the read cache is divided into two parts, the read cache closed user adopt smaller data block unit and set-associative mapping program with the HASH search algorithm, such a design can improve the temporal locality of caching system. The read cache closed data adopt larger data block unit and fully associative mapping programs, such a design can improve the spatial locality of caching system. At the same time, by modeling read cache and determine the reasonable proportion between reserved part and dynamic part according to theoretical analysis. The storage system performance was greatly improved according to the above design.Fault handling is the performance source of the disk arrays. The foundation of fault handling operation and destaging operation is reconstructure, so, the the research mainly focus on reconstructure operation. The recovery path in new software RAID architure is different from main code path and the concept of generating matrix is introduced in reconstruction operation, now, the reconstruction operation can use the formula of matrix according to the matrix attributions by reprenting stripe using generating matrix, pseudo-inverse reconstruction formula can be quickly obtained, Thereby, enhancing the system performance of reconstruction. Comparing to reconstruction algorithm highly depended on the specific RAID erasure codes and coding layout, the matrix reconstruction algorithm have a stronger ability.Finally, we will summarize the work done by this paper and discuss future research direction is likely to continue.