SEA-SSD: A Storage Engine Assisted SSD With Application-Coupled Simulation Platform

A storage engine assisted solid-state drive (SEA-SSD) has been proposed to improve the write performance of the SSDs for database applications. Since the SSD write performance bottleneck is caused by the garbage collection, the motivation of this work is to reduce the SSD garbage collection overhead by identifying, clustering and aggregating the data with similar activity in the same block of the NAND flash memory. This paper proposes the SEA-SSD to exploit the upstream information in the storage engine (SE) of database to classify the data pattern and predict the data activity. Hint information like the buffer pool/redo log size, page clean/dirty status, and frequency and access intervals are passed from the SE to the SSD controller. In order to co-design and optimize the database SE and SSD controller, an application-coupled platform has been developed and utilized for the evaluation of the SEA-SSD, which accelerates the simulation speed over 20-times and reduces the implementation complexity. Moreover, another two platforms are developed as well for the system-level optimizations of the SSDs, which are not specifically limited to the SEA-SSD. With the proposed SEA-SSD, maximum 24% performance improvement, 16% energy consumption reduction and 19% lifetime extension are achieved from the experimental results.