期刊
EXPERT SYSTEMS WITH APPLICATIONS
卷 221, 期 -, 页码 -出版社
PERGAMON-ELSEVIER SCIENCE LTD
DOI: 10.1016/j.eswa.2023.119697
关键词
Parameter Optimization; Redis; Genetic Algorithm; Deep Learning
In this study, we propose the DARK tuning system to improve the performance of Redis, an in-memory key-value store. The tuning was performed by classifying knobs related to persistence methods. We also propose the Cross-GA method, which alternates between prediction and alignment using Genetic Algorithm to improve throughput and latency simultaneously. Through performance evaluations using Memtier-benchmark, the proposed method achieved up to 39.8% improvement in throughput and 71.3% improvement in latency compared to the existing configuration.
Database Management System (DBMS) runs the server with numerous knobs having different roles, which are set in the configuration file, and the performance of the DBMS can easily depend on the configuration. It could derive greater performance by finding the appropriate values of knobs, but there is a limitation for an expert who seeks to ascertain the most effective parameter combination directly due to the diversity of knobs and the wide range of values that each parameter could have. In this study, we propose the DARK tuning system to improve the performance of Redis, an in-memory key-value store. Since the two persistence methods provided in Redis are not designed to operate simultaneously, the tuning was performed by classifying knobs related to persistence methods. Also, we propose the Cross-GA, a method of conducting the prediction and alignment alternately of Genetic Algorithm to improve throughput and latency simultaneously. To verify the effectiveness of the proposed method, we carried out performance evaluations through Memtier-benchmark, a benchmark program for in-memory databases. As a result of performing Redis knobs tuning via DARK, the optimal configuration was derived to improve throughput up to 39.8% and latency up to 71.3% compared to the existing configuration.
作者
我是这篇论文的作者
点击您的名字以认领此论文并将其添加到您的个人资料中。
推荐
暂无数据