Real Time and Performance Management Techniques in SSD Storage Systems

Flash-based storage systems offer high density, robustness, and reliability for embedded applications; however the physical nature of flash memory means that there are limitations to its usage in high reliability applications. To increase the reliability of flash-based storage systems, several RAID mechanisms have been proposed. However, these mechanisms permit the recovery of data onto a new replacement device when a particular device in the array reaches its endurance limit and they need regular garbage collection to efficiently manage free resources. These present concerns with response time as when a garbage collector or a device replacement is underway, the flash memory cannot be used by the application layer for an uncertain period of time. This non-determinism in terms of response time is problematic in high reliability systems that require real-time guarantees. Existing solutions to garbage collection only consider single flash chip but ignore architectures where multiple flash memories are used in a storage system such as RAID. Traditional replacement mechanisms based on magnetic storage mediums do not suit specifications of flash memory. The aim of this thesis is to improve the reliability of the SSD RAID mechanisms by providing guaranteed access time for hard real-time embedded applications. Investigating the hypothesis, a number of novel mechanisms were proposed with the goal of enhancing data reliability in an SSD array. Two novel mechanisms solve the non-determinism problem caused by garbage collection without disturbing the reliability mechanism unlike existing techniques. The third mechanism is device replacement techniques for replacing elements in the array, increasing system dependability by providing continuous system availability with higher I/O performance for hard real-time embedded applications. A global flash translation layer with novel garbage collection mechanisms, on-line device replacement techniques, and their associated controllers are implemented on our FPGA SSD RAID controller. Contrary to traditional approaches, a dynamic preemptive cleaning mechanism adopts a dynamic cleaning feature which does not disturb the reliability mechanism. In addition to this the garbage collection aware RAID mechanism is introduced to improve the maximum response time of the system further. On-line device replacement techniques address limitations of the device replacement and thus provide more deterministic response times. The reliability, real-time and performance of these mechanisms via trace-driven simulator for number of synthetic and realistic traces are also evaluated. The contribution of this thesis is as follows: the presentation of novel mechanisms that enable the real-time support for RAID techniques in SSD devices, the development of a number of mechanisms that enhance the performance and reliability of flash-based storages, the implementation of these controllers, and the provision of a complete test bed for investigating these behaviours.