Reliability Management Techniques in SSD Storage Systems

Solid State Devices (SSDs) are becoming ubiquitous in many embedded devices due to their features such as no moving parts, shock/temperature resistance and low power consumption. The reliability, performance, lifespan, and verification of these devices is an increasing issue in many application domains. Redundant Array of Independent Disk (RAID) architectures have previously been used to increase reliability of magnetic disk devices. Recent works have proposed re-use of RAID architectures to address these issues in SSDs, however these have an inherent problem as all flash memory chips wear out at the same rate due to even distribution of write operations. Existing solutions partly solve this problem using uneven parity distribution across the array, but suffer age-variation problems under random and sequential writes thereby decreasing the reliability of the array as well as increasing the cost. The aim of this thesis is to enhance the use of RAID mechanisms in SSD storage systems, and to do so in a reliable, and efficient manner. In this thesis the two novel mechanisms are explored that enhance data reliability in a SSD array regardless of I/O workload characteristics. The first mechanism solves the age convergence problem in RAID systems and quickly achieves steady-state age convergence. The second mechanism reduces page writes, thereby increasing the lifespan of each element in the array. The SSD controller ( ANFS ) architecture and the associated RAID architecture (flash-RAID) are presented. The embedded Flash Translation Layer (FTL), RAID controller, SSD low level controller, and specialized host interface are developed on an FPGA in synthesizable Verilog. In these architectures the concept of a forced random write is introduced which is used to solving the age distribution problem of pure sequential writes. It further employs a log-structured approach to control on-chip wear-levelling, and a power-fail reliability mechanism. In addition to this a new flash management framework is presented that increases the performance of SSD storage systems in the exploitation of both multi-chip parallelism and out-of-order execution. The contribution of this thesis can be summarised as follows: the presentation of new algorithms that successfully enable the efficient and reliable RAID techniques in SSD devices, the development of a number of techniques that enhance the performance and reliability of flash-based file systems, the implementation of a controller in synthesizable Verilog that employs these techniques, and the provision of a complete test bed supporting the experiments.