An Efficient Design and Implementation of Multi-level Cache for Database Systems

Flash-based solid state device(SSD) is making deep inroads into enterprise database applications due to its faster data access. The capacity and performance characteristics of SSD make it well-suited for use as a second-level buffer cache. In this paper, we propose a SSD-based multilevel buffer scheme, called flash-aware second-level cache(FASC), where SSD serves as an extension of the DRAM buffer. Our goal is to reduce the number of disk accesses by caching the pages evicted from DRAM in the SSD, thereby enhancing the performance of database systems. For this purpose, a cost-aware main memory replacement policy is proposed, which can efficiently reduce the cost of page evictions. To take full advantage of the SSD, a block-based data management policy is designed to save the memory overheads, as well as reducing the write amplification of flash memory. To identify the hot pages for providing great performance benefits, a memory-efficient replacement policy is proposed for the SSD. Moreover, we also present a light-weight recovery policy, which is used to recover the data cached in the SSD in case of system crash. We implement a prototype based on PostgreSQL and evaluate the performance of FASC. The experimental results show that FASC achieves significant performance improvements.