面向应用性能管理系统的运行负载预测

Operational Load Forecasting for Application Performance Management Systems

摘要:

在应用性能管理系统中,系统未来的负载情况对运维调度有重要的指导意义。在云计算环境下,弹性伸缩计算能力为调整系统规模提供了可能,根据系统将来的负载情况可以提前做出相应的调整：可以在负载加重前扩展好集群,保证服务质量；在负载降低之后若预测一定时间内没有负载加重的情况,则可以及时缩减集群规模,降低企业运营成本。在金融领域,ARIMA模型是常用的时序预测模型,但其应用需要人工介入分析时序的平稳性,调参过程过于复杂。近年来神经网络技术的发展带动了人工更智能技术的发展,本论文设计并测试了ANN、RNN、GRU、LSTM等神经网络的负载预测的效果。实验结果表明LSTM网络预测精准且表现稳定,是系统负载预测的理想模型。

Abstract:

In the application performance management system, the future load situation of the system has important guiding significance for the operation and maintenance scheduling. In the cloud computing environment, the elastic scaling computing capability provides the possibility to adjust the system scale. According to the future load conditions of the system, it can be adjusted in advance: You can expand the cluster before the load is increased to ensure the quality of service; after the load is reduced, If no load is expected to increase in a certain period of time, the scale of the cluster can be reduced in time and the operating cost of the enterprise can be reduced. In the financial field, the ARIMA model is a commonly used time-series prediction model, but its application requires manual intervention to analyze the temporal stability, and the adjustment process is too complicated. In recent years, the development of neural network technology has led to the development of artificially more intelligent technologies. This paper designed and tested the effect of neural network load prediction such as ANN, RNN, GRU, and LSTM. Experimental results show that the LSTM network is accurate and stable in performance and is an ideal model for system load prediction.