AC-key: Adaptive caching for LSM-based key-value stores

Fenggang Wu, Ming Hong Yang, Baoquan Zhang, David H.C. Du

Research output: Chapter in Book/Report/Conference proceedingConference contribution

2 Scopus citations

Abstract

Read performance of LSM-tree-based Key-Value Stores suffers from serious read amplification caused by the leveled structure used to improve write performance. Caching is one of the main techniques to improve the performance of read operations. Designing an efficient caching algorithm is challenging because the leveled structure obscures the cost and benefit of caching a particular key, and the trade-off between point lookup and range query operations further complicates the cache replacement decisions. We propose AC-Key, an Adaptive Caching enabled KeyValue Store to address these challenges. AC-Key manages three different caching components, namely key-value cache, key-pointer cache, and block cache, and adjust their sizes according to the workload. AC-Key leverages a novel caching efficiency factor to capture the heterogeneity of the caching costs and benefits of cached entries. We implement AC-Key by modifying RocksDB. The evaluation results show that the performance of AC-Key is higher than that of RocksDB in various workloads and is even better than the best offline fix-sized caching scheme in phase-change workloads.

Original languageEnglish (US)
Title of host publicationProceedings of the 2020 USENIX Annual Technical Conference, ATC 2020
PublisherUSENIX Association
Pages603-615
Number of pages13
ISBN (Electronic)9781939133144
StatePublished - 2020
Event2020 USENIX Annual Technical Conference, ATC 2020 - Virtual, Online
Duration: Jul 15 2020Jul 17 2020

Publication series

NameProceedings of the 2020 USENIX Annual Technical Conference, ATC 2020

Conference

Conference2020 USENIX Annual Technical Conference, ATC 2020
CityVirtual, Online
Period7/15/207/17/20

Bibliographical note

Funding Information:
We thank the anonymous ATC reviewers and our anonymous shepherds for their feedback. This work was partially supported by NSF I/UCRC Center Research in Intelligent Storage and the following NSF awards 1439622, 1525617, and 1812537.

Publisher Copyright:
Copyright © Proc. of the 2020 USENIX Annual Technical Conference, ATC 2020. All rights reserved.

Fingerprint Dive into the research topics of 'AC-key: Adaptive caching for LSM-based key-value stores'. Together they form a unique fingerprint.

Cite this