Lifespan-based garbage collection to improve SSD's reliability and performance

Abstract

Given the “out-of-place” nature of write mode, Solid State Drive (SSD) has to execute a garbage collection (GC) to reclaim the space of invalidated data and free flash blocks for new written data. However, the GC can incur the movements of valid data inside SSD, resulting in the problems such as write latency, write amplification, etc. The GC overheads of SSD depend not only on the current write pattern but also on how data have been already placed in SSD. If data can be classified by their lifetime when written to SSD, and the data with similar lifetime are written into the same flash block, we can select the block with the shortest expected lifespan as the victim block. Ideally, the victim block contains few valid data, implying there are few valid pages that need to be migrated during the GC process with greatly reduced overhead. In this paper, we propose and implement a GC method, called Lifespan-based GC, for SSDs based on the data lifetime, together with the I/O distribution and the dynamic flash memory allocation. The results show that for those applications with the data lifetime characteristics, the lifespan-based GC can effectively reduce the amount of valid pages migrated in GC process, thereby minimizing the write latency and write amplification caused by GC.