| 摘 要: 针对Kubernetes默认水平伸缩策略在高并发场景下因间歇时间而导致集群规模无法及时扩展,进而易引发集群性能下降甚至宕机的问题,提出了一种基于加权变分减法双向长短期记忆网络模型(INFO-VMD-SABOBiLSTM,IVS-BiLSTM)的容器水平伸缩策略。该策略通过将负载预测值输入Pod水平伸缩器(Horizontal Pod Autoscaler,HPA)进行主动扩容,提升了集群对负载变化的感知能力。实验结果表明,所提的IVS-BiLSTM混合预测模型在平均绝对误差(MAE)、均方根误差(RMSE)和平均绝对百分比误差(MAPE)3项指标上分别减少到LSTM的44.29%、44.34%和43.03%,并且较其他主流算法在精度上有显著提升;改进的容器水平伸缩策略在相同负载条件下能够提前扩容,证明了该策略在生产中的可行性。 |
| 关键词: Kubernetes 弹性伸缩 预测 INFO SABO HPA |
|
中图分类号: TP319
文献标识码: A
|
| 基金项目: 江苏省重点研发计划(BE2020084-4) |
|
| Researchon Container Horizontal Scaling Strategy Based on Hybrid Prediction Model |
|
ZHAO Yifan1, JIANG Lingyun1,2
|
(1. School of Communication and Information Engineering, Nanjing University of Posts and Telecommunications, Nanjing 210003, China;
2. Internet of Things Research Institute, Nanjing University of Posts and Telecommunications, Nanjing 210003, China)
1222014521@njupt.edu.cn; jiangly@njupt.edu.cn
|
| Abstract: To address the issue of performance degradation or even cluster downtime caused by untimely scaling due to intermittent periods under high-concurrency scenarios in Kubernetes’default horizontal scaling policy, this study proposes a container horizontal scaling strategy based on a Weighted Variational Mode DecompositionBidirectional Long Shor-t Term Memory hybrid prediction model (VMD-SABO-BiLSTM, IVS-BiLSTM). By feeding load prediction values into the Horizontal Pod Autoscaler (HPA) for proactive scaling, this strategy enhances the cluster’s responsiveness to workload fluctuations.Experimental results show that the proposed IVS-BiLSTM hybrid prediction model reduces the three indicators of Mean Absolute Error (MAE), Root Mean Square Error (RMSE), and Mean Absolute Percentage Error (MAPE) to 44.29% , 44.34% , and 43.03% of those of LSTM, respectively, and significantly improves the accuracy compared with other mainstream algorithms.The improved container horizontal scaling strategy enables earlier scaling under identical load conditions, validating its feasibility for production
environments. |
| Keywords: Kubernetes elastic scaling prediction INFO SABO HPA |
