Data Temperature-Aware Bloom Filters for Flash-Based Storage
Wei Li, Tianyun Zhang, Dafang Zhang, Yupeng Hu, Kun Xie, Ting Zhu · 2019
Hot data identification techniques for flash memory have attracted a plethora of research interests because of its crucial impact on performance and life span of SSD. Our proposed technique, Data Temperature-Aware Bloom Filters (DTABF), is suitable for the requirements of both the Buffer Management and the Flash Translation Layer in SSD. DTABF divides the access period of LPAs into n cycles. Thus, DTABF can record the access frequency of LPAs with a round-robin fashion in each cycle by combining one Counting Bloom filter and n Bloom filters. The access pattern of each LPA, indicated by the data access frequency, recency, and access frequency changes over time, is presented with n different data temperatures in n cycles. Specifically, we employ a bijective function to characterize the data temperature of each LPA. Based on the data temperature identified by DTABF, the data stored in LPAs, which presenting the similar access pattern, can be gathered in the same flash block in SSD. Meanwhile, those data occasionally becoming cold in the buffer can be effectively identified to avoid being meaninglessly replaced from the buffer. Analytical and experimental results show that DTABF can help to improve hit rate, write performance and alleviate write amplification, while achieving lower memory cost and computational complexity.