Kubernetes容器云的弹性伸缩方法

doi:10.16180/j.cnki.issn1007-7820.2025.03.005

摘要/Abstract

摘要：

Kubernetes容器云是当前流行的云计算技术,其默认的弹性伸缩方法HPA(Horizontal Pod Autoscaler)能对云原生应用进行横向扩缩容。但该方法存在以下问题:基于单一负载指标,使其难以适用于多样化云原生应用;基于当前负载进行弹性伸缩,使扩缩容过程具有明显的滞后性;基于滑动时间窗口算法进行弹性缩容,使缩容过程缓慢易造成系统资源浪费。针对上述问题,文中提出一种改进的弹性伸缩方法。设计一种动态加权融合算法将多种负载指标融合为综合负载因子,全面反映云原生应用的综合负载。提出CEEMDAN(Complete Ensemble Empirical Mode Decomposition with Adaptive Noise)-ARIMA(Autoregressive Integrated Moving Average Model)预测模型,基于该模型的预测负载值实现预先弹性伸缩以应对突发流量。提出快速缩容与滑动时间窗口相结合的方法,在确保应用服务质量的基础上减少系统资源浪费。实验结果表明,相较于HPA机制,改进的弹性伸缩方法在应对首次突发流量时的平均响应时间缩短了336.55%,流量结束后系统资源占用减少了50%,再次遇到突发流量时能迅速扩容,平均响应时间缩短66.83%。

关键词: 云计算, 容器云, Kubernetes, HPA, 弹性伸缩, 时间序列, 滑动窗口, 加权融合, 负载预测

Abstract:

Kubernetes container cloud is currently a popular cloud computing technology, and its default elastic scaling method HPA(Horizontal Pod Autoscaler) can horizontally expand and shrink cloud native applications.However, this method is based on a single load index, which is difficult to apply to diversified cloud-native applications. In addition, the method performs elastic expansion based on the current load, so that the process of expansion and contraction has obvious hysteresis. This method is based on the sliding time window algorithm for elastic shrinkage, which is slow and easy to waste system resources.To solve these problems, an improved elastic stretching method is proposed in this paper. A dynamic weighted fusion algorithm is designed to fuse multiple load indicators into comprehensive load factors, which can fully reflect the comprehensive load of cloud native applications.CEEMDAN(Complete Ensemble Empirical Mode Decomposition with Adaptive Noise)-ARIMA(Autoregressive Integrated Moving Average Model) prediction model is proposed, and elastic expansion is realized in advance to cope with the burst traffic based on the predicted load value of the model.A method combining rapid capacity reduction and sliding time window is proposed to reduce system resource waste on the basis of ensuring application service quality.Experimental results show that compared with the HPA mechanism, the improved elastic scaling method shortens the average response time by 336.55% when dealing with the first burst traffic, reduces system resource usage by 50% after the traffic ends, and can quickly expand the capacity when encountering burst traffic again, with an average response time shortened by 66.83%.

Key words: cloud computing, container cloud, Kubernetes, HPA, auto scaling, time series, sliding window, weighted fusion, load forecasting

中图分类号:

TP301.6

李佳颖, 杨泽民, 宋哲代, 朱金荣. Kubernetes容器云的弹性伸缩方法[J]. 电子科技, 2025, 38(3): 32-39.

LI Jiaying, YANG Zemin, SONG Zhedai, ZHU Jinrong. Research on Elastic Scaling Method of Kubernetes Container Cloud[J]. Electronic Science and Technology, 2025, 38(3): 32-39.

图/表 9

图1

图2

图3

图4

表1

图5

表2

图6

图7

参考文献 25

[1]	Yegulalp S. Why you should use docker and containers[J]. Information World, 2023, 37(1):1-3.
[2]	Carmen C. Kubernetes as a standard container orchestrator-A bibliometric analysis[J]. Journal of Grid Computing, 2022, 20(4):1-4.
[3]	董子奇, 刘淇, 高原, 等. 基于容器技术的微服务部署研究[J]. 信息技术与标准化, 2023, 64(Z1):93-98.
	Dong Ziqi, Liu Qi, Gao Yuan, et al. Research of microservice deployment based on container technology[J]. Information Technology and Standardization, 2023, 64(Z1):93-98.
[4]	Khaldoun S, Sohail A, Naveed A, et al. A survey of Kubernetes scheduling algorithms[J]. Journal of Cloud Computing, 2023, 12(1):1-7.
[5]	Bisson S. Using microsoft azure's prometheus monitoring with Kubernetes[J]. Information World, 2023, 37(3):1-4.
[6]	Huo Q Z, Li C Y, Li S N, et al. High concurrency response strategy based on Kubernetes horizontal pod autoscaler[J]. Journal of Physics:Conference Series, 2023, 2451(1):1-7.
[7]	Guo J F, Huang C L, Hou J L. A scalable computing resources system for remote sensing big data processing using GeoPySpark based on spark on K8s[J]. Remote Sensing, 2022, 14(3):1-5.
[8]	郑浩, 李宁, 杨小涛. 自适应负载指标权值的负载均衡算法[J]. 计算机工程与设计, 2019, 40(3):623-626,643.
	Zheng Hao, Li Ning, Yang Xiaotao. Load balancing algorithm with adaptive load index weight[J]. Computer Engineering and Design, 2019, 40(3):623-626,643.
[9]	邰滢滢, 薄旭, 庞影, 等. 一种运用模糊综合评判的实时服务器负载均衡算法[J]. 小型微型计算机系统, 2019, 40(1):93-97.
	Tai Yingying, Bo Xu, Pang Ying, et al. Load balancing algorithm of real-time server based on fuzzy comprehensive judgement[J]. Journal of Chinese Computer System, 2019, 40(1):93-97.
[10]	Premathas S, Radu C. Predicting locally manageable resource failures of high availability clusters[J]. Software:Practice and Experience, 2022, 52(10):1-6.
[11]	Zhen L T, Zhang L M, Yang T X, et al. Simultaneous prediction for multiple source-loads based sliding time window and convolutional neural network[J]. Energy Reports, 2022, 8(1):1-6.
[12]	李松远. QoS感知的服务资源调度与优化研究[D]. 北京: 北京邮电大学,2021:2-13.
	Li Songyuan. QoS-aware service resource scheduling and optimization[D]. Beijing: Beijing University of Posts and Telecommunications,2021:2-13.
[13]	沐磊, 李洪赭, 李赛飞. 一种改进的Kubernetes弹性伸缩策略[J]. 计算机与数字工程, 2022, 50(2):327-331,372.
	Mu Lei, Li Hongzhe, Li Saifei. An improved Kubernetes elastic scaling strategy[J]. Computer and Digital Engineering, 2022, 50(2):327-331,372.
[14]	龚俊宇. 基于云原生架构的即时消息系统的设计与实现[D]. 西安: 西安电子科技大学,2021:4-17.
	Gong Junyu. Design and implementation of instant mssaging system based on cloud native architecture[D]. Xi'an: Xidian University,2021:4-17.
[15]	Qian J J, Lin H F. A forest fire identification system based on weighted fusion algorithm[J]. Forests, 2022, 13(8):1-5.
[16]	Kewen W, Gang Y, Yongzheng L, et al. A parameterized iterative synchrosqueezing transform for the analysis of noise contaminated non-stationary signals[J]. Measurement, 2023, 21(6):1-5.
[17]	Sangjo L, Sangyeol L, Koichi M. Change point test for structural vector autoregressive model via independent component analysis[J]. Journal of Statistical Computation and Simulation, 2023, 93(5):1-6.
[18]	余琼芳, 牛冬阳. 基于LSTM网络的矿山压力时空混合预测[J]. 电子科技, 2023, 36(2):67-72.
	Yu Qiongfang, Niu Dongyang. Mixed prediction of mine pressure time and space based on LSTM network[J]. Electronic Science and Technology, 2023, 36(2):67-72.
[19]	Kumar P S, Ranjan M T, Khushi A. Forecasting road accidental deaths in India: An explicit comparison between ARIMA and exponential smoothing method[J]. International Journal of Injury Control and Safety Promotion, 2023, 30(4):1-6.
[20]	Jing T. Quality evaluation of preschool video games based on RNN[J]. Entertainment Computing, 2023, 47(6):1-5.
[21]	Nhat Minh D, Myungsik Y. Deep learning-based autoscaling using bidirectional long short-term memory for Kubernetes[J]. Applied Sciences, 2021, 11(9):1-5.
[22]	Leonardo G D, Matteo L, Fabio S, et al. Improved maximum likelihood estimation of ARMA Models[J]. Lobachevskii Journal of Mathematics, 2022, 43(9):1-4.
[23]	聂巧平, 孙淇铭. 时间序列模型的Ljung-Box检验的发展与应用综述[J]. 统计与管理, 2022, 37(4):101-107.
	Nie Qiaoping, Sun Qiming. A review of the development and application of Ljung-Box test for ime series models[J]. Statistics and Management, 2022, 37(4):101-107.
[24]	Guanling Z, Lisheng H, Lu L, et al. Prediction of industrial network security situation based on noise reduction using EMD[J]. Mobile Information Systems, 2022, 22(3):1-4.
[25]	梁荣华. 基于资源负载预测的容器云弹性伸缩策略研究[D]. 桂林: 桂林理工大学,2022:5-20.
	Liang Ronghua. Research on auto scaling strategy of container cloud based on resource load prediction[D]. Guilin: Guilin University of Technology,2022:5-20.

主机名	操作系统	CPU核心数	内存/GB
Master	CentOS7	2	2
Node₁	CentOS7	4	4
Node₂	CentOS7	4	4
本机	Ubuntu18	8	16

预测模型	MAE	MSE	RMSE
ARIMA	1.152 4	2.572 5	1.603 9
LSTM	1.543 8	3.040 9	1.975 3
EMD-ARIMA	0.954 7	1.880 6	1.371 4
CEEMDAN-ARIMA	0.790 3	1.446 9	1.202 9