Integrating de-duplication and write for increased performance and endurance of Solid-State Drives

摘要

Modern NAND Flash-based Solid-State Drives (SSD) presents low latency, high throughput, low power consumption and solid-state reliability improvements comparing to traditional magnetic-disk based Hard Disk Drives (HDD). However, due to NAND Flash memory cell characteristics, update-in-place is impossible. Instead, the Flash software layer allocates new storage space whenever data is written, even if it is a slightly modified version of already stored data, e.g., a slightly modified file. Consequently, a logical overwrite entails extensive writing with resulting latency, power and chip endurance costs. Noting that reads are much faster than writes and that stale (“overwritten”) data is seldom erased until a much later time, we present an SSD architecture with integrated de-duplication at the Flash software layer. Our concept is to use de-duplication tools to manage write and read operations in SSD so as to reduce redundant writes caused by data overwrite. Read operation incur a penalty in order to reconstruct the stored data from multiple data versions. Our proposed architecture''s benefit grows in multi-block writes and in write-mostly applications. It moreover offers the side benefit of incremental backup, which enables to restore previous versions of the stored data.