《计算机应用》唯一官方网站 ›› 2022, Vol. 42 ›› Issue (12): 3763-3768.DOI: 10.11772/j.issn.1001-9081.2021101725

• 数据科学与技术 • 上一篇    

面向单记录的混合负载下物化视图异步增量维护任务生成

孙洋洋1(), 姚俊萍1, 李晓军1, 范守祥2, 王自维3   

  1. 1.火箭军工程大学 301教研室, 西安 710025
    2.中国人民解放66133部队, 北京 100043
    3.火箭军工程大学 205教研室, 西安 710025
  • 收稿日期:2021-10-09 修回日期:2021-12-21 接受日期:2021-12-22 发布日期:2022-01-04 出版日期:2022-12-10
  • 通讯作者: 孙洋洋
  • 作者简介:姚俊萍(1978—),女,陕西渭南人,副教授,博士,主要研究方向:智能信息处理
    李晓军(1981—),男,河北秦皇岛人,副教授,博士,主要研究方向:数据质量、信息系统、人因工程
    范守祥(1986—),男,辽宁丹东人,助理工程师,硕士,主要研究方向:智能信息处理
    王自维(1996—),男,甘肃平凉人,硕士研究生,主要研究方向:检测与诊断、自动化装置。

Materialized view asynchronous incremental maintenance task generation under hybrid transaction/analytical processing for single record

Yangyang SUN1(), Junping YAO1, Xiaojun LI1, Shouxiang FAN2, Ziwei WANG3   

  1. 1.Teaching and Research Office 301,Rocket Force University of Engineering,Xi’an Shaanxi 710025,China
    2.PLA 66133 Troop,Beijing 100043,China
    3.Teaching and Research Office 205,Rocket Force University of Engineering,Xi’an Shaanxi 710025,China
  • Received:2021-10-09 Revised:2021-12-21 Accepted:2021-12-22 Online:2022-01-04 Published:2022-12-10
  • Contact: Yangyang SUN
  • About author:YAO Junping,born in 1978, Ph. D., associate professor. Her research interests include intelligent information processing.
    LI Xiaojun,born in 1981, Ph. D., associate professor. His research interests include data quality, information system, human factors engineering.
    FAN Shouxiang, born in 1986, M. S, associate engineer. His research interests include intelligent information processing.
    WANG Ziwei, born in 1996, M. S. candidate. His research interests include detection and diagnosis, automation device.

摘要:

针对已有的混合负载(HTAP)下物化视图异步增量维护任务生成算法主要面向多记录,无法面向单记录生成HTAP物化视图异步增量维护任务,导致磁盘IO开销的增加,进而降低HTAP物化视图异步增量维护性能的问题,提出面向单记录的HTAP物化视图异步增量维护任务的生成方法。首先,建立面向单记录的HTAP物化视图异步增量维护任务生成的效益模型;然后,基于Q-learning设计面向单记录的HTAP物化视图异步增量维护任务的生成算法。实验结果表明,所提算法在实现面向单记录生成HTAP物化视图异步增量维护任务的基础上,将平均每秒读写操作次数(IOPS)、平均CPU利用率(2核)和平均CPU利用率(4核)至少分别降低了8.49次、1.85个百分点和0.97个百分点。

关键词: 混合负载, 物化视图维护, 强化学习, Q-learning, 异步增量维护, 单记录维护任务生成

Abstract:

Existing materialized view asynchronous incremental maintenance task generation algorithms under Hybrid Transaction/Analytical Processing (HTAP) are mainly used for multiple records and unable to generate materialized view asynchronous incremental maintenance task under HTAP for single record, which results in the increase of disk IO overhead and the performance degradation of materialized view asynchronous incremental maintenance under HTAP. Therefore, a materialized view asynchronous incremental maintenance task generation method under HTAP for single record was proposed. Firstly, the benefit model of materialized view asynchronous incremental maintenance task generation under HTAP for single record was established. Then, the materialized view asynchronous incremental maintenance task generation under HTAP for single record algorithm was designed on the basis of Q-learning. Experimental results show that materialized view asynchronous incremental maintenance task generation under HTAP for single record is realized by the proposed algorithm, and the proposed algorithm decreases the average IOPS (Input/output Operations Per Second), average CPU utilization (2-core) and average CPU utilization (4-core) at least by 8.49 times, 1.85 percentage points and 0.97 percentage points respectively.

Key words: Hybrid Transaction/Analytical Processing (HTAP), materialized view maintenance, reinforcement learning, Q-learning, asynchronous incremental maintenance, maintenance task generation for single record

中图分类号: