Page 154 - 《软件学报》2025年第12期

荣垂田 等: 多目标深度强化学习驱动的数据库系统参数优化技术 5535

[22] Liu X, Liu SY, Zhuang YK, Gao Y. Explainable reinforcement learning: Basic problems exploration and method survey. Ruan Jian Xue
Bao/Journal of Software, 2023, 34(5): 2300–2316 (in Chinese with English abstract). http://www.jos.org.cn/1000-9825/6485.htm [doi: 10.
13328/j.cnki.jos.006485]
[23] Hu TM, Luo B. PA2D-MORL: Pareto ascent directional decomposition based multi-objective reinforcement learning. In: Proc. of the
38th AAAI Conf. on Artificial Intelligence. Vancouver: AAAI Press, 2024. 12547–12555. [doi: 10.1609/aaai.v38i11.29148]
[24] Nguyen TT, Nguyen ND, Vamplew P, Nahavandi S, Dazeley R, Lim CP. A multi-objective deep reinforcement learning framework.
Engineering Applications of Artificial Intelligence, 2020, 96: 103915. [doi: 10.1016/j.engappai.2020.103915]
[25] Reymond M, Bargiacchi E, Nowé A. Pareto conditioned networks. In: Proc. of the 21st Int’l Conf. on Autonomous Agents and
Multiagent Systems. Auckland: Int’l Foundation for Autonomous Agents and Multiagent Systems, 2022. 1110–1118.
[26] Castanet N, Sigaud O, Lamprier S. Stein variational goal generation for adaptive exploration in multi-goal reinforcement learning. In:
Proc. of the 40th Int’l Conf. on Machine Learning. Honolulu: JMLR, 2023. 151.
[27] Hu TM, Luo B, Yang CH, Huang TW. MO-MIX: Multi-objective multi-agent cooperative decision-making with deep reinforcement
learning. IEEE Trans. on Pattern Analysis and Machine Intelligence, 2023, 45(10): 12098–12112. [doi: 10.1109/TPAMI.2023.3283537]
[28] Alegre LN, Bazzan ALC, Roijers DM, Nowé A, da Silva BC. Sample-efficient multi-objective learning via generalized policy
improvement prioritization. In: Proc. of the 2023 Int’l Conf. on Autonomous Agents and Multiagent Systems. London: Int’l Foundation
for Autonomous Agents and Multiagent Systems, 2023. 2003–2012.
[29] Watson LT, Haftka RT. Modern homotopy methods in optimization. Computer Methods in Applied Mechanics and Engineering, 1989,
74(3): 289–305. [doi: 10.1016/0045-7825(89)90053-4]

附中文参考文献:
[4] 李国良, 周煊赫. 轩辕: AI 原生数据库系统. 软件学报, 2020, 31(3): 831–844. http://www.jos.org.cn/1000-9825/5899.htm [doi: 10.13328/
j.cnki.jos.005899]
[22] 刘潇, 刘书洋, 庄韫恺, 高阳. 强化学习可解释性基础问题探索和方法综述. 软件学报, 2023, 34(5): 2300–2316. http://www.jos.org.cn/
1000-9825/6485.htm [doi: 10.13328/j.cnki.jos.006485]

附录 A

本文中实验部分使用的 MySQL 的参数和 PostgreSQL 的参数见表 A1 和表 A2.

表 A1 MySQL 推荐参数 Knobs 调优详情

+
MySQL CDBTune ON-MODDPG
Knobs 初始值
默认值 TPC-C RW RO WO TPC-C RW RO WO
binlog_format Statement Mixed Mixed Mixed Row Row Mixed Mixed Row Mixed
innodb_buffer_pool_size 128 MB 4 GB 6 GB 8 GB 8 GB 8 GB 8 GB 8 GB 8 GB 8 GB
innodb_log_files_in_group 2 2 8 12 1 2 12 12 11 2
innodb_log_file_size 48 MB 0.5 GB 3.3 GB 0.1 GB 15 GB 39 GB 8.3 GB 8.3 GB 3.9 GB 32 GB
innodb_read_io_threads 4 12 3 1 1 31 52 64 1 28
binlog_cache_size 32 KB 32 MB 0.7 GB 0.4 GB 4 KB 4 KB 1 GB 4 KB 0.7 GB 0.8 GB
innodb_buffer_pool_instances 8 8 8 1 1 8 14 2 6 25
max_binlog_cache_size 18.4 EB 4 GB 4 GB 1.5 GB 4.7 GB 0.2 GB 3.5 GB 3 GB 4.4 GB 5 GB
binlog_checksum CRC32 None None CRC32 CRC32 CRC32 None CRC32 CRC32 None
innodb_purge_threads 1 1 25 15 25 29 4 21 32 11
max_binlog_size 1 GB 1 GB 0.4 GB 0.3 GB 69 MB 0.2 GB 0.2 GB 0.6 GB 0.2 GB 0.3 GB
innodb_write_io_threads 4 12 52 1 2 64 64 64 1 8
skip_name_resolve Off On On On Off On Off Off On On
innodb_file_per_table On On Off Off Off On Off On On Off
table_open_cache 2 000 512 524 288 524 288 397 935 524 288 442 241 380 371 16 715 216 686
max_connections 151 1 600 50 000 7 144 7 910 47 208 50 000 48 731 20 928 20 860