Latency-Aware Request Scheduling for NAND Flash-based SSDs

Abstract

This paper proposes a new latency-aware request scheduling method for NAND flash-based solid state disks (SSDs). Requests can be categorized as user requests or system requests. User requests are generated by the host, and system requests are generated by the SSD controller due to garbage collection, which is a function of flash translation layer (FTL). The priority of system requests is usually higher than user requests because the garbage collection should normally be done in advance for subsequent write requests. Thus, the response time of user requests significantly increases when system requests are generated. To solve this problem, the proposed method reduces the response time of user requests when system requests are generated by scheduling user requests and system requests using different queues connected to same Flash chip. Experimental results and analysis on the proposed method show that the response time of user requests can be reduced by up to 27% by efficiently using the interaction between software queues and the FTL at the software level, thereby efficiently exploit multi-chip parallelism.