iLife: Safely Extending Lifetime for Memory-Oriented SSD

Abstract

Currently, the roles of SSDs have been diversified significantly. Apart for storage purposes, users can also utilize flash drives to cache hot data or buffering incoming writes to achieve high throughput. In this case, the endurance study of Solid State Drives (SSDs) has become an increasingly important topic as users rely on such research to reliably avoid early retirement of their old drives in laptops or machines in data centers. However, current manufacturers adopt Program/Erase (P/E) cycle-based life indicators to estimate the remaining lifetime of an SSD, which is often too conservative due to longtime data retention concerns. In this paper, we begin by analyzing the inaccuracy and other potential issues of existing P/E cycle-based life indicators for SSDs under memory-oriented workloads. To construct an accurate life indicator, we conduct a more extensive and sophisticated experiment to wear out eight SSDs from four different vendors. By monitoring the device status extracted from the standard S.M.A.R.T.attributes in the experiment, we make eleven interesting findings on the relationship between the device lifetime and different types of errors observed. Based on our unique findings, we propose iLife, a pattern-based life indicator for SSDs under memory-oriented workloads. Our evaluation of iLife on a different set of six SSDs shows that iLife is a more accurate life indicator, with an average of 14.2% higher accuracy on lifetime estimation and up to 21.2% safe lifetime extension compared to existing P/E cycle-based indicators.