Design And Optimization Of Solid State Storage System Based On Nand Flash

Limited by the large rotating seek delay created by mechanical device, the traditionalhard disk is difficult to meet the demand for high performance storage systems. Solid statestorage have high read and write speed, low power consumption, and is increasingly beingapplied to storage system. PCIe(Peripheral Component Interconnection Express) solidstate drive is usually used in enterprise-class applications that require solid state storagesystem with high performance and large capacity, but each implementation is incompatible.NVMe (Non-volatile Memory Express) protocol defined a communication protocolspecification and command set between operating system and non-volatile subsystem, andimproved the compatibility of system. The command defined in NVMe support forcollaborative optimization of operating system and solid state drive.To satisfy the demand of large capacity and high performance storage systems forenterprise-class applications, a complete solid state storage system solution was designed.The storage system includes a solid-state storage devices drive and solid state storage card,using NVMe to implement data transmission between host and solid state drive. Through amultiple command queue mechanism, this system can meet the highly concurrent accesscharacteristics of enterprise applications load. The solid-state memory card is detaileddesigned and divided into NVMe protocol stack modules, flash conversion layer modules,solid state storage card platform module. To solve the problem cause by random write andthe capacity of mapping table, we propose a sub-double page hybrid mapping strategy, anda scheme to cache the mapping table to a small buffer. We design a complete FTL (Flashtranslation layer) including the buffer management strategy and garbage collectionalgorithm, to work in conjunction with the sub-double page hybrid mapping strategy.This solid state storage system solution can adapt to enterprise-class applications withmuch random write load and demand for large capacity. By using the optimized flashtranslation layer, we can improve write performance, while the smaller mapping tablespace system programs can keep scalability.