Exploring Cascaded Write Amplification in LSM-Tree Based Key-Value Stores with Solid-State Disks

摘要

Log-structured merge tree (i.e., LSM-tree) based key-value stores, which are widely used in big-data applications, provide high performance. NAND Flash-based Solid state disks (i.e., SSDs) become the popular devices that alternative to hard disk drives (i.e., HDDs) because of the high performance and low power consumption. LSM-tree KV stores with SSDs are deployed in large-scale storage systems to achieve high performance in the cloud. However, the host-side write amplification in LSM-tree KV stores and the device-side write amplification in SSDs can impair the overall system performance. The former that is attributed to compaction has a burn on I/O bandwidth between two sides. The latter, which results from garbage collection in NAND Flash memory, blocks user I/Os from the host and degrades the SSD performance. In this study, we explored cascaded write amplification in LSM-tree KV stores with SSDs. Our primary goal is to comprehensively study two-level cascaded write amplification on the host and device sides. We quantitatively analyze the impact of two-level write amplification on overall performance. This work reveals details of cascaded write amplification and helps developers in the software stacks of LSM-tree based KV stores with SSDs.