Rethinking Logging, Checkpoints, and Recovery for High-Performance Storage Engines

Abstract

For decades, ARIES has been the standard for logging and recovery in database systems. ARIES offers important features like support for arbitrary workloads, fuzzy checkpoints, and transparent index recovery. Nevertheless, many modern inmemory database systems use more lightweight approaches that have less overhead and better multi-core scalability but only work well for the in-memory setting. Recently, a new class of high-performance storage engines has emerged, which exploit fast SSDs to achieve performance close to pure in-memory systems but also allow out-of-memory workloads. For these systems, ARIES is too slow whereas inmemory logging proposals are not applicable. In this work, we propose a new logging and recovery design that supports incremental and fuzzy checkpointing, index recovery, out-of-memory workloads, and low-latency transaction commits. Our continuous checkpointing algorithm guarantees bounded recovery time. Using per-thread logging with minimal synchronization, our implementation achieves near-linear scalability on multi-core CPUs. We implemented and evaluated these techniques in our LeanStore storage engine. For working sets that fit in main memory, we achieve performance close to that of an in-memory approach, even with logging, checkpointing, and dirty page writing enabled. For the out-of-memory scenario, we outperform a state-of-the-art ARIES implementation by a factor of two.

DecLog: Decentralized Logging in Non-Volatile Memory for Time Series Database Systems

“…Also, a passive batch commit strategy is utilized to realize the synchronization between multi-thread loggings. Haubenschild et al [26] extend GSN based logging and propose a Remote Flush Avoidance (RFA) protocol to reduce the dependency between logs. Taurus [72] features parallel logging and employs an LSN vector to track transaction dependencies.…”

“…Some recent studies [15,16,19,23,26,28,37,67] utilize NVM to improve logging performance. To ensure data consistency after recovery, the log order in NVM must be consistent with the transaction commitment order.…”

“…Therefore, decentralized logging approaches are proposed to replace centralized ones to improve the logging performance. Two main categories of approaches exist: Epoch-based logging [18,61,79] and global sequence number (GSN) based logging [26,67]. Although these proposals avoid concurrent contention, additional protocol and instruction overheads are introduced.…”

The Impact of Privacy Regulations on DB Systems

“…A research opportunity presents in using such ideas in a more modern logging scheme, such as [Haubenschild et al 2020], which low-latency logging and bounded recovery features can be exploited in this context, allowing for fast recovery of important data. Another area to be explored is anomaly detection through log analysis, in which possible violations can be detected.…”

A Performance Evaluation of DRAM Access for In-Memory Databases

“…In main memory systems, using pointers can not only save considerable storage space, which is important for expensive DRAM, but can also simplify handling variable length fields [4], [13], [25], such as memory allocation and cache alignment. This fully benefits from the random access performance in the main memory.…”