FASTer FTL for Enterprise-Class Flash Memory SSDs

• Authors: Sang-Phil Lim;Sang-Won Lee;Bongki Moon
• Venue: SNAPI '10 Proceedings of the 2010 International Workshop on Storage Network Architecture and Parallel I/Os
• Year: 2010

For the past decade, numerous methods have been proposed for the design of a flash translation layer (FTL), which is the core engine of flash memory drives that critically determines the performance of the drives. In this paper, we revisit one of the popular FTL schemes called FAST, and augment it with new optimization techniques aiming particularly at online transaction processing (OLTP) workloads. As flash memory solid state drives (SSDs) are increasingly adopted for large-scale enterprise-class storage systems, it is important to develop an FTL that can deal with OLTP workloads in a scalable manner, which are characterized by a large number of small, random and skewed IO operations. With the proposed optimization methods such as giving a second chance to valid pages and isolating cold ones, the enhanced FTL, called FASTer, outperforms FAST considerably. In our experiments, FASTer reduced average elapsed time by more than 30 percent, and minimized the fluctuation of response time drastically. Overall, the performance of FASTer was comparable to a page mapping FTL, which tends to consume much more DRAM space to store a large mapping table. This will make the FASTer FTL one of the most cost effective and scalable FTL schemes for OLTP workloads.