ICAP, a new flash wear-leveling algorithm inspired by locality

Abstract

In this paper, we extend a trivial wear-leveling algorithm to reduce the management overhead of leveraging the number of write operations over all data blocks. By exploiting access locality, most parts of a mapping table of a flash translation layer (FTL) keep unchanged since wear leveling is adopted only when a block tends to be unreliable. Because the mapping table is changed rarely, we can reduce the mounting time by record most portion of the mapping table on the flash drive. Furthermore, we can give short codes to the data blocks that erased frequently (because the locality) when recording the erase counts in a log structure. We observe that the algorithm can extend the lifetime of a flash drive and reduce mounting time in most situations.

Can We Group Storage? Statistical Techniques to Identify Predictive Groupings in Storage System Accesses

“…In a large system for long-term storage, the effect of these infrequent accesses can accumulate to be a large drain on power [Wildani and Miller 2010]. Recent studies have investigated using temporal locality [Lo et al 2014] or spatial locality [Wu and He 2012] to more intelligently manage the flash translation in solid-state devices. The grouping methodologies we discuss have been tested for a variety of use cases including selectively prefetching fingerprints into a cache [Wildani et al 2013], which may be a useful starting point for a generalized flash grouping technique.…”