Flash memory management with cooperation, adaptation and assistance

Abstract

Flash memory is ubiquitous for data storage today. However, existing strategies to manage a flash device is insufficient due to inherent characteristics of flash and various workloads it is serving. This thesis has proposed three novel approaches for flash memory management. The first approach is on the inter-module cooperation inside the management firmware. Observational Wear Leveling has been designed with the cooperation between wear leveling and address mapping. The second approach is to be workload-adaptive. ADPAT and TreeFTL have been devised. They are for address mapping and RAM buffer management, respectively, and can adapt to different workloads for data access. The third approach is to seek assistance from the OS for flash management. OS-Assisted Wear leveling (SAW) has been developed. The OS of SAW provides hints for the management firmware to utilize. Experiments have been conducted to these proposals, and improvements are significant on device lifetime and access performance, respectively.