The uFLIP benchmark revisited for evaluating SSDs

Abstract

Solid-state drives (SSDs) provide higher bandwidth and lower access latency than hard disk drives (HDDs) and thus have been rapidly replacing HDDs. SSD manufacturers, which produce different types of SSDs, do not open their SSD architecture and control firmware in detail. To find the best SSD for a given application, it is necessary to compare and analyze the performances of different kinds of SSDs. A benchmark is widely used to analyze the storage devices performance. However, most existing benchmarks are mainly targeted for HDDs. The uFLIP (understanding Flash IO Patterns) is a benchmark proposed to consider flash devices' characteristics. In this paper, we first exploit the uFLIP benchmark to analyze and to evaluate the performances of different types of SSDs. Through extensive experiments, we have identified common SSD characteristics: (1) the input/output (IO) size and queue depth affect the parallelism that is an inherent characteristic in SSDs; (2) the size of SSDs access space considerably affects the SSDs performance. Then, via a series of experiments, we verify whether the uFLIP benchmark could correctly characterize applications by its micro-benchmarks. The results reveal that the performance of the uFLIP micro-benchmarks, produced to have features similar to that of a target application, is quite different from the performance of the target application.