No abstract

The log-structured merge tree (LSM-tree) has become an essential component in many key-value systems and expanded its scope to full-fledged database engines (e.g., MyRocks). In the database landscape, vendors face growing customer demands for real-time analytic solutions to handle hybrid transactional/analytical processing (HTAP) workloads that pose significant challenges. Among the challenges is IO amplification that drives system designers to rethink write-optimized engines to survive HTAP loads. This paper follows the same philosophy, reexamines LSM-trees used for database systems, and rethinks IO amplification under HTAP loads to shed some light on practical remedies for upcoming challenges. We propose two practical techniques to alleviate IO amplification: 1) aligned compaction for reducing write amplification, 2) snapshot filters for reducing read amplification. The two techniques are lightweight (i.e., near-zero resource consumption) and are compatible with state-of-the-art methods. We integrated our techniques into RocksDB and demonstrated that the modified RocksDB exhibits reduced IO amplification under HTAP workloads with negligible resource consumption.

Hybrid transactional/analytical processing (HTAP) would overload database systems. To alleviate performance interference between transactions and analytics, recent research pursues the potential of in-storage processing (ISP) using commodity computational storage devices (CSDs). However, in-storage query processing faces technical challenges in HTAP environments. Continuously updated data versions pose two hurdles: (1) data items keep changing, and (2) finding visible data versions incurs excessive data access in CSDs. Such access patterns dominate the cost of query processing, which may hinder the active deployment of CSDs. This paper addresses the core issues by proposing an a nalyt i c offloa d e ngine (AIDE) that transforms engine-specific query execution logic into vendor-neutral computation through a canonical interface. At the core of AIDE are the canonical representation of vendor-specific data and the separate management of data locators. It enables any CSD to execute vendor-neutral operations on canonical tuples with separate indexes, regardless of host databases. To eliminate excessive data access, we prescreen the indexes before offloading; thus, host-side prescreening can obviate the need for running costly version searching in CSDs and boost analytics. We implemented our prototype for PostgreSQL and MyRocks, demonstrating that AIDE supports efficient ISP for two databases using the same FPGA logic. Evaluation results show that AIDE improves query latency up to 42× on PostgreSQL and 34× on MyRocks.