Timestamp-based hot/cold data identification scheme for solid state drives

Abstract

Flash memory is a non-volatile memory that has been widely used as a storage medium for various mobile devices, consumer electronics, and data centers due to its natures of lightweight, high performance, low power consumption, and shock resistance. However, flash memory requires erasing before it can be overwritten. Compared with other operations, the erase operation is the most time-consuming. In addition, flash memory can only endure a limited number of erasures. Out-place-update is adopted to hide the overhead incurred by erase operations. The space occupied by obsolete data are reclaimed during garbage collection. Garbage collection reclaims free space by migrating valid data from the victim block to another free flash block, and then erasing the victim block. To improve the performance of garbage collection and extend the lifetime of the storage device, we propose a new data separation scheme, referred to as the Enhance Dynamic Clustering (EDC) scheme. By this scheme, data are dynamically classified and clustered together according to their data lifetimes. Experiment results showed that the EDC scheme significantly improved the performance of garbage collection, compared with various schemes. The number of erase operations and extra write operations performed during garbage collection could be greatly reduced even under various types of host workloads.