Research On Object-based Storage System Architecture And QoS Strategy

With the development of large-scale computer system, existing storage systems have faced more and more challenges, including higher I/O speed, larger storage capacity, expansibility and usability. Traditional network storage systems (NAS, SAN) have many limitations because of their architectures, which make them unfit for large-scale system. In this situation, a new conception–OBS (Object-Based Storage System) has been put forward in recent years. OBS adopts object conception to manage data and have powerful intelligence to fulfill advanced functions. Profiting from the ability of OSD, the heavy tasks of OBS can be distributed to several OSDs, leaving only management tasks to system server. This architecture owns good expansibility, security and scalability. The system performance will almost keep linear increase with the extension of OBS scale.The most important content of this paper is putting forward an object-based high performance storage system. The key properties are as follows: firstly, it makes the best of the objects'attributes with adaptable services and pre-fetch strategy to improve accessing performance; secondly, every OSD has its own processor and storage operating system, which can both avoid system bottleneck and offer asynchronous I/O; thirdly, using SRAID (Similar RAID) architecture to improve the expansibility and usability of the system. The experiments results show that this storage system evidently improve random I/O performance for small blocks of data, compared with traditional storage systems. Also, the multi-thread and asynchronous I/O modes improve the expansibility of the storage system.The second part of this paper is researching the QoS (Quality of Service) guarantee mechanism of OBS. Through designing QoS attribute page and QoS architecture, to satisfy the storage network and storage devices'QoS requirements.