《计算机应用》唯一官方网站 ›› 2023, Vol. 43 ›› Issue (6): 1884-1892.DOI: 10.11772/j.issn.1001-9081.2022050722
所属专题: 先进计算
收稿日期:
2022-05-20
修回日期:
2022-08-04
接受日期:
2022-08-08
发布日期:
2023-06-08
出版日期:
2023-06-10
通讯作者:
刘曙光
作者简介:
方和平(1997—),男,重庆人,硕士研究生,CCF会员,主要研究方向:深度强化学习、绿色数据中心基金资助:
Heping FANG1,2, Shuguang LIU1(), Yongyi RAN3, Kunhua ZHONG1
Received:
2022-05-20
Revised:
2022-08-04
Accepted:
2022-08-08
Online:
2023-06-08
Published:
2023-06-10
Contact:
Shuguang LIU
About author:
FANG Heping, born in 1997, M. S. candidate. His research interests include deep reinforcement learning, green data center.Supported by:
摘要:
多数据中心任务调度策略的目的是把计算任务分配到各个数据中心的不同服务器上,以促进资源利用率和能效的提升,为此提出了基于深度强化学习的多数据中心一体化调度策略。所提策略分为数据中心选择和数据中心内部任务分配两个阶段。在多数据中心选择阶段,整合算力资源以提高总体资源利用率,首先采用具有优先经验回放的深度Q网络(PER-DQN)在以数据中心为节点的网络中获取到达各个数据中心的通信路径;然后计算资源使用成本和网络通信成本,并依据这两个成本之和最小的原则选择最优的数据中心。在数据中心内部任务分配阶段,首先在所选数据中心内部,划分计算任务并遵循先到先服务(FCFS)原则将任务添加到调度队列中;然后结合计算设备状态和环境温度,采用基于双深度Q网络(Double DQN)的任务分配算法获得最优分配策略,以选择服务器执行计算任务,避免热点的产生,并降低制冷设备的能耗。实验结果表明,基于PER-DQN的数据中心选择算法相较于计算资源优先(CRF)、最短路径优先(SPF)路径选择方法的平均总成本分别下降了3.6%、10.0%;基于Double DQN的任务部署算法相较于较轮询调度(RR)、贪心调度(Greedy)算法的平均电源使用效率(PUE)分别下降了2.5%、1.7%。可见,所提策略能够有效降低总成本和数据中心能耗,实现多数据中心的高效运行。
中图分类号:
方和平, 刘曙光, 冉泳屹, 钟坤华. 基于深度强化学习的多数据中心一体化调度优化[J]. 计算机应用, 2023, 43(6): 1884-1892.
Heping FANG, Shuguang LIU, Yongyi RAN, Kunhua ZHONG. Integrated scheduling optimization of multiple data centers based on deep reinforcement learning[J]. Journal of Computer Applications, 2023, 43(6): 1884-1892.
资源 | 费用 |
---|---|
每个CPU核(core) | |
每GB内存 | |
每GB磁盘 | |
每Mb通信带宽 |
表1 资源费用示例
Tab.1 Resource price examples
资源 | 费用 |
---|---|
每个CPU核(core) | |
每GB内存 | |
每GB磁盘 | |
每Mb通信带宽 |
参数 | 设定值 |
---|---|
训练情节数 | 80 000 |
折扣因子 | 0.99 |
内存容量 | 5 000 |
学习率 | 0.001 |
1 | |
0.001 | |
0.000 166 5 | |
批量尺寸 | 128 |
目标网络更新周期 | 70 |
隐藏层数 | 2 |
表2 多数据中心任务调度模型参数设置
Tab.2 Multiple data center task scheduling model parameters setting
参数 | 设定值 |
---|---|
训练情节数 | 80 000 |
折扣因子 | 0.99 |
内存容量 | 5 000 |
学习率 | 0.001 |
1 | |
0.001 | |
0.000 166 5 | |
批量尺寸 | 128 |
目标网络更新周期 | 70 |
隐藏层数 | 2 |
数据中心 | 每个CPU核数 单位时间价格 | RAM/GB | Disk/GB | 带宽/Mb |
---|---|---|---|---|
1 | 0.116 4 | 1.581 | 0.058 | 0.334 |
2 | 0.112 2 | 1.585 | 0.061 | 0.456 |
3 | 0.106 3 | 1.565 | 0.069 | 0.385 |
4 | 0.110 2 | 1.567 | 0.063 | 0.489 |
5 | 0.119 0 | 1.605 | 0.060 | 0.399 |
6 | 0.101 0 | 1.705 | 0.072 | 0.367 |
7 | 0.125 0 | 1.495 | 0.052 | 0.347 |
8 | 0.105 0 | 1.475 | 0.035 | 0.397 |
9 | 0.111 0 | 1.464 | 0.051 | 0.343 |
10 | 0.122 0 | 1.450 | 0.045 | 0.385 |
11 | 0.117 0 | 1.563 | 0.055 | 0.356 |
表3 数据中心资源单位时间价格
Tab.3 Prices of data center resources in unit time
数据中心 | 每个CPU核数 单位时间价格 | RAM/GB | Disk/GB | 带宽/Mb |
---|---|---|---|---|
1 | 0.116 4 | 1.581 | 0.058 | 0.334 |
2 | 0.112 2 | 1.585 | 0.061 | 0.456 |
3 | 0.106 3 | 1.565 | 0.069 | 0.385 |
4 | 0.110 2 | 1.567 | 0.063 | 0.489 |
5 | 0.119 0 | 1.605 | 0.060 | 0.399 |
6 | 0.101 0 | 1.705 | 0.072 | 0.367 |
7 | 0.125 0 | 1.495 | 0.052 | 0.347 |
8 | 0.105 0 | 1.475 | 0.035 | 0.397 |
9 | 0.111 0 | 1.464 | 0.051 | 0.343 |
10 | 0.122 0 | 1.450 | 0.045 | 0.385 |
11 | 0.117 0 | 1.563 | 0.055 | 0.356 |
参数 | 设定值 |
---|---|
训练情节数 | 9 000 |
折扣因子 | 0.99 |
内存容量 | 5 000 |
学习率 | 0.001 |
0.2 | |
0.001 | |
0.000 019 9 | |
批量尺寸 | 128 |
目标网络更新周期 | 30 |
隐藏层数 | 3 |
表4 数据中心内部任务部署模型参数设置
Tab.4 Data center internal task deployment model parameters settings
参数 | 设定值 |
---|---|
训练情节数 | 9 000 |
折扣因子 | 0.99 |
内存容量 | 5 000 |
学习率 | 0.001 |
0.2 | |
0.001 | |
0.000 019 9 | |
批量尺寸 | 128 |
目标网络更新周期 | 30 |
隐藏层数 | 3 |
算法 | 平均奖励 | 平均PUE | 平均过载数 |
---|---|---|---|
RR | 15.03±2.84 | 2.36±0.015 | 57±16.63 |
Greedy | 15.95±2.82 | 2.34±2.360 | 38±20.88 |
Double DQN | 17.45±2.77 | 2.30±0.017 | 33±16.63 |
表5 实验结果数据
Tab.5 Experimental results data
算法 | 平均奖励 | 平均PUE | 平均过载数 |
---|---|---|---|
RR | 15.03±2.84 | 2.36±0.015 | 57±16.63 |
Greedy | 15.95±2.82 | 2.34±2.360 | 38±20.88 |
Double DQN | 17.45±2.77 | 2.30±0.017 | 33±16.63 |
1 | LI Q R, PENG Z P, CUI D L, et al. Two-stage selection of distributed data centers based on deep reinforcement learning [J]. Cluster Computing, 2022, 25(4): 2699-2714. 10.1007/s10586-021-03525-8 |
2 | JIN Y B, GAO Y, QIAN Z Z, et al. Workload-aware scheduling across geo-distributed data centers[C]// Proceedings of the 15th IEEE International Conference on Trust, Security and Privacy in Computing and Communications/ 10th IEEE International Conference on Big Data Science and Engineering/ 14th IEEE International Symposium on Parallel and Distributed Processing with Applications. Piscataway: IEEE, 2016: 1455-1462. 10.1109/trustcom.2016.0228 |
3 | CAO Z W, ZHOU X, HU H, et al. Towards a systematic survey for carbon neutral data centers [J]. IEEE Communications Surveys and Tutorials, 2022, 24(2): 895-936. 10.1109/comst.2022.3161275 |
4 | PAKBAZNIA E, GHASEMAZAR M, PEDRAM M. Temperature-aware dynamic resource provisioning in a power-optimized datacenter [C]// Proceedings of the 2010 Design, Automation and Test in Europe Conference and Exhibition. Piscataway: IEEE, 2010: 124-129. 10.1109/date.2010.5457223 |
5 | XU J L, PALANISAMY B. Cost-aware resource management for federated clouds using resource sharing contracts [C]// Proceedings of the IEEE 10th International Conference on Cloud Computing. Piscataway: IEEE, 2017: 238-245. 10.1109/cloud.2017.38 |
6 | SHEN H Y, CHEN L H. A resource usage intensity aware load balancing method for virtual machine migration in cloud datacenters[J]. IEEE Transactions on Cloud Computing, 2020, 8(1): 17-31. 10.1109/tcc.2017.2737628 |
7 | GE Y J, WEI G Y. GA-based task scheduler for the cloud computing systems[C]// Proceedings of the 2010 International Conference on Web Information Systems and Mining. Piscataway: IEEE, 2010: 181-186. 10.1109/wism.2010.87 |
8 | ETTIKYALA K, LATHA Y V. Rank based efficient task scheduler for cloud computing[C]// Proceedings of the 2016 International Conference on Data Mining and Advanced Computing. Piscataway: IEEE, 2016: 343-346. 10.1109/sapience.2016.7684151 |
9 | YI D L, ZHOU X, WEN Y G, et al. Toward efficient compute-intensive job allocation for green data centers: a deep reinforcement learning approach [C]// Proceedings of the IEEE 39th International Conference on Distributed Computing Systems. Piscataway: IEEE, 2019: 634-644. 10.1109/icdcs.2019.00069 |
10 | RAN Y Y, HU H, ZHOU X, et al. DeepEE: joint optimization of job scheduling and cooling control for data center energy efficiency using deep reinforcement learning [C]// Proceedings of the IEEE 39th International Conference on Distributed Computing Systems. Piscataway: IEEE, 2019: 645-655. 10.1109/icdcs.2019.00070 |
11 | CHENG M X, LI J, BOGDAN P, et al. H₂O-Cloud: a resource and quality of service-aware task scheduling framework for warehouse-scale data centers[J]. IEEE Transactions on Computer-Aided Design of Integrated Circuits and Systems, 2020, 39(10): 2925-2937. 10.1109/tcad.2019.2930575 |
12 | 祝家钰,肖丹,王飞. 云计算下负载均衡的多维QoS约束任务调度机制[J]. 计算机工程与应用, 2013, 49(9):85-89. 10.3778/j.issn.1002-8331.1110-0676 |
ZHU J Y, XIAO D, WANG F. Multi-dimensional QoS constrained scheduling mechanism based on load balancing for cloud computing[J]. Computer Engineering and Applications, 2013, 49(9):85-89. 10.3778/j.issn.1002-8331.1110-0676 | |
13 | 左利云,曹志波. 云计算中调度问题研究综述[J]. 计算机应用研究, 2012, 29(11):4023-4027. 10.3969/j.issn.1001-3695.2012.11.005 |
ZUO L Y, CAO Z B. Review of scheduling research in cloud computing[J]. Application Research of Computers, 2012, 29(11): 4023-4027. 10.3969/j.issn.1001-3695.2012.11.005 | |
14 | ZHAO J, LI H X, WU C, et al. Dynamic pricing and profit maximization for the cloud with geo-distributed data centers [C]// Proceedings of the 2014 IEEE Conference on Computer Communications. Piscataway: IEEE, 2014: 118-126. 10.1109/infocom.2014.6847931 |
15 | ALZHOURI F, AGARWAL A, LIU Y. Maximizing cloud revenue using dynamic pricing of multiple class virtual machines [J]. IEEE Transactions on Cloud Computing, 2021, 9(2): 682-695. 10.1109/tcc.2018.2878023 |
16 | RAN Y Y, YANG J, ZHANG S B, et al. Dynamic IaaS computing resource provisioning strategy with QoS constraint[J]. IEEE Transactions on Services Computing, 2017, 10(2): 190-202. 10.1109/tsc.2015.2464212 |
17 | GU L, ZENG D Z, GUO S, et al. A general communication cost optimization framework for big data stream processing in geo-distributed data centers[J]. IEEE Transactions on Computers, 2016, 65(1): 19-29. 10.1109/tc.2015.2417566 |
18 | HILDA J, CHANDRASEKARAN S. Cost and time economical planning algorithm for scientific workflows in cloud computing [J]. Future Internet, 2021, 13(10): No.263. 10.3390/fi13100263 |
19 | WANG W, YAN Y, ZHANG L M, et al. Collaborative sparse coding for multiview action recognition[J]. IEEE MultiMedia, 2016, 23(4): 80-87. 10.1109/mmul.2016.69 |
20 | MNIH V, KAVUKCUOGLU K, SILVER D, et al. Playing Atari with deep reinforcement learning [EB/OL]. (2013-12-19) [2022-04-20].. 10.1038/nature14236 |
21 | van HASSELT H, GUEZ A, SILVER D. Deep reinforcement learning with double Q-learning[C]// Proceedings of the 30th AAAI Conference on Artificial Intelligence. Palo Alto, CA: AAAI Press, 2016: 2094-2100. 10.1609/aaai.v30i1.10295 |
22 | SCHAUL T, QUAN J, ANTONOGLOU I, et al. Prioritized experience replay [EB/OL]. (2016-02-25) [2022-04-20].. |
23 | GÓMEZ-MARTÍN C, VEGA-RODRÍGUEZ M A, GONZÁLEZ-SÁNCHEZ J L. Performance and energy aware scheduling simulator for HPC: evaluating different resource selection methods[J]. Concurrency and Computation: Practice and Experience, 2015, 27(17): 5436-5459. 10.1002/cpe.3607 |
24 | LI Q R, PENG Z P, CUI D L, et al. Data center selection based on reinforcement learning [C]// Proceedings of the 4th International Conference on Cloud Computing and Internet of Things. Piscataway: IEEE, 2019: 14-19. 10.1109/cciot48581.2019.8980333 |
25 | The Hebrew University — Experimental Systems Lab. Logs of real parallel workloads from production systems [EB/OL]. (2019-09-24) [2022-05-09].. |
26 | PATERIA S, SUBAGDJA B, TAN A H, et al. Hierarchical reinforcement learning: a comprehensive survey[J]. ACM Computing Surveys, 2021, 54(5): No.109. 10.1145/3453160 |
27 | BACON P L, HARB J, PRECUP D. The option-critic architecture[C]// Proceedings of the 31st AAAI Conference on Artificial Intelligence. Palo Alto, CA: AAAI Press, 2017: 1726-1734. 10.1609/aaai.v31i1.10916 |
28 | KULKARNI T D, NARASIMHAN K R, SAEEDI A, et al. Hierarchical deep reinforcement learning: integrating temporal abstraction and intrinsic motivation[C]// Proceedings of the 30th International Conference on Neural Information Processing Systems. Red Hook, NY: Curran Associates Inc., 2016: 3682-3690. 10.48550/arXiv.1604.06057 |
[1] | 周毅, 高华, 田永谌. 基于裁剪优化和策略指导的近端策略优化算法[J]. 《计算机应用》唯一官方网站, 2024, 44(8): 2334-2341. |
[2] | 马天, 席润韬, 吕佳豪, 曾奕杰, 杨嘉怡, 张杰慧. 基于深度强化学习的移动机器人三维路径规划方法[J]. 《计算机应用》唯一官方网站, 2024, 44(7): 2055-2064. |
[3] | 赵晓焱, 韩威, 张俊娜, 袁培燕. 基于异步深度强化学习的车联网协作卸载策略[J]. 《计算机应用》唯一官方网站, 2024, 44(5): 1501-1510. |
[4] | 唐睿, 庞川林, 张睿智, 刘川, 岳士博. D2D通信增强的蜂窝网络中基于DDPG的资源分配[J]. 《计算机应用》唯一官方网站, 2024, 44(5): 1562-1569. |
[5] | 秦鑫彤, 宋政育, 侯天为, 王飞越, 孙昕, 黎伟. 基于自适应p持续的移动自组网信道接入和资源分配算法[J]. 《计算机应用》唯一官方网站, 2024, 44(3): 863-868. |
[6] | 李源潮, 陶重犇, 王琛. 基于最大熵深度强化学习的双足机器人步态控制方法[J]. 《计算机应用》唯一官方网站, 2024, 44(2): 445-451. |
[7] | 邓辅秦, 官桧锋, 谭朝恩, 付兰慧, 王宏民, 林天麟, 张建民. 基于请求与应答通信机制和局部注意力机制的多机器人强化学习路径规划方法[J]. 《计算机应用》唯一官方网站, 2024, 44(2): 432-438. |
[8] | 余家宸, 杨晔. 基于裁剪近端策略优化算法的软机械臂不规则物体抓取[J]. 《计算机应用》唯一官方网站, 2024, 44(11): 3629-3638. |
[9] | 龙杰, 谢良, 徐海蛟. 集成的深度强化学习投资组合模型[J]. 《计算机应用》唯一官方网站, 2024, 44(1): 300-310. |
[10] | 尚绍法, 蒋林, 李远成, 朱筠. 异构平台下卷积神经网络推理模型自适应划分和调度方法[J]. 《计算机应用》唯一官方网站, 2023, 43(9): 2828-2835. |
[11] | 王昱, 任田君, 范子琳. 基于引导Minimax-DDQN的无人机空战机动决策[J]. 《计算机应用》唯一官方网站, 2023, 43(8): 2636-2643. |
[12] | 王子腾, 于亚新, 夏子芳, 乔佳琪. 融合好奇心和策略蒸馏的稀疏奖励探索机制[J]. 《计算机应用》唯一官方网站, 2023, 43(7): 2082-2090. |
[13] | 李校林, 江雨桑. 无人机辅助移动边缘计算中的任务卸载算法[J]. 《计算机应用》唯一官方网站, 2023, 43(6): 1893-1899. |
[14] | 黄晓辉, 杨凯铭, 凌嘉壕. 基于共享注意力的多智能体强化学习订单派送[J]. 《计算机应用》唯一官方网站, 2023, 43(5): 1620-1624. |
[15] | 曹腾飞, 刘延亮, 王晓英. 基于改进深度强化学习的边缘计算服务卸载算法[J]. 《计算机应用》唯一官方网站, 2023, 43(5): 1543-1550. |
阅读次数 | ||||||
全文 |
|
|||||
摘要 |
|
|||||