Research On Key Technologies Of RAID5-based Disk Array Built By Flash-based SSDs

With the rapid development of network and digital device, the modern world gradually goes into the world of data. The storage system is facing a great challenge. Besides the huge demand on the storage’s scale, there is also a higher request for latency, bandwidth, reliability and energy consumption. Because of the limited internal mechanism of the traditional hard disk drive, IO bandwidth and latency has become the major bottleneck of large scale storage systems. Flash-based SSD(Solid State Drive) has been widely used in enterprise market, due to its high performance, non-volatile and low energy consumption. However, Flash memory has a drawback of limited lifetime, and reliability should be taken into consideration when applying Flash memory into large scale storage system. RAID mechanism can be used to construct RAID storage system based on flash memory to increase the capacity, elevate the performance and improve the reliability, which provides an efficient way to build storage systems with large scale, high performance and high reliability. The paper mainly focus on the research of RAID5-based SSD array. According to the characteristics of RAID5, we optimize the internal structure of SSD, and improve the control logic of RAID5 based on the feature of Flash memory. Through which we prolong the lifetime of RAID5-based SSD arrays. The contributions of this dissertation are summarized as below:First, because of the limited lifetime of Flash memory, we propose Parity Aware-SSD, which can be aware of parity sent from RAID5 controller. This structure place the parity information generated by RAID5 controller in SSD reasonably. It reduces the write operations generated by SSD controller and prolong the lifetime of SSD. Inside the SSD controller, we propose a weight-based replace streagy for write cache and optimize the address mapping and wear leveling of Flash Translation Layer.Second, we improve the structure of RAID5 controller. Focusing on the data locality of applications, we add write cache in the RAID5 controller to reduce write operations of data. To solve the problem of frequent parity update in RAID5 system, we set partial parity cache and data buffer. Data buffer receives data replaced by write cache, and partial parity cache stores the corresponding partial parity of updated data, which reduces the read and write operations of SSD when updating parities.Finally, we implement an RAID5-based SSD array simulator, by which we test the performance of proposed PA-SSD and RAID5 controller using several traces. The results show that, PA-SSD reduces 26% parity write, 17% total write and 32% page movement on average. Optimized RAID5 controller reduces 66% writes on SSD and 44% write latency on average.