快閃記憶體儲存系統耐用度提昇之高效能管理策略

Abstract

NAND 型快閃記憶體已經被廣泛地使用在各種不同的儲存系統，且它的應用已經超過它原有的設計。近年來，成本已經成為快閃記憶體技術發展的主要驅動力，然而也同時在耐用度上帶來嚴重的挑戰。本論文提出一個只需少量主記憶體且不需要對目前的管理方法做大量修改就能有效提昇快閃記憶體耐用度的高效靜態抹除機制，其目的為使用一個有效率的方法來盡可能地平均分散區塊抹除的分佈。而當考量到低成本的快閃記憶體系統時，本論文提出一個提交式的管理機制來解決系統耐用度與可靠性的問題，並且只需要少量的主記憶體就可以達成，此管理機制不只提昇系統的耐用度與可靠性，同時也改善系統在讀寫、初始化及當機回復的效能。本論文最後提出一個驅動程式層的設計，在不修改任何既有快閃記憶體管理機制的情況下，就能提昇低成本快閃記憶體儲存系統的可靠性，進而延長系統的耐用度，其目的為有效地更正大量資料錯誤與隨機資料錯誤。最後，透過一連串的實驗來評估所提方法的性能，並展示所提出設計的可行性。

NAND flash memory is widely adopted in various storage systems, and its applications have grown beyond its original designs. In recent years, cost has become a major driving force in the development of the flash memory technology. Because of that, future products are now facing serious challenges on endurance requirements. In this dissertation, an efficient static wear leveling design is proposed to enhance the endurance of flash memory with limited overhead and without many modifications to popular implementation designs. It aims at distributing erases over blocks of flash memory as evenly as possible with an efficient implementation. When low-cost flash-memory storage systems are considered, we propose a commitment-based management strategy to improve their reliability so as to extend their endurance with limited main-memory requirements. It is not only to enhance the endurance and reliability of systems, but also to improve the system performance on read and write requests, initialization, and crash recovery. This dissertation is concluded by the exploring of a driver-layer design to further enhance reliability of low-cost flash-memory storage systems so as to extend their endurance without any modification to existing flash management designs. It aims at correcting burst errors and random errors effectively and efficiently. The capabilities of the proposed approaches were evaluated by a series of experiments to demonstrate the effectiveness of the designs.