基于时序卷积网络的云服务器性能预测模型

目前基于深度学习的主机性能预测模型大部分缺乏普适性，实验数据缺乏公正性，无法准确预测能耗或性能峰值点且时间开销较大.为解决这些问题，文章提出了一种基于改进时序卷积网络的云服务器性能预测模型(ATCN模型).该模型将CPU利用率作为主机过载的衡量标准，利用多维性能指标构建N+1维能耗向量，建立输入向量与预测标准之间的关系；调整TCN中的卷积核大小并不断增大扩张因子，实现长期记忆效果.基于阿里云开源数据集的实验结果表明:ATCN模型具有强自适应性，在不同硬件配置和资源使用情况下，预测准确率和效率方面比LSTM模型提升大约20%.

A Prediction Model of Cloud Server Performance Based on Temporal Convolutional Network

Most of the current deep learning-based models for host performance prediction do not have universal applicability, cannot produce fair experimental data or accurately predict the energy consumption or performance peak point, and consume too much time. In order to solve these problems, an adaptive performance prediction mo-del based on Temporal Convolutional Network (ATCN Model) is proposed. In this model, CPU utilization is regarded as the measurement standard for host overload, N+1-dimension energy consumption vector is constructed with multi-dimensional performance index, and the relationship between the input vector and prediction standard is established. The convolution kernel size in TCN is adjusted and the expansion factor is increased continuously to achieve a long-term memory effect. The experimental results based on Alibaba cloud open-source data set show that the ATCN model has strong self-adaptability. Under different hardware configurations and resource usages, the prediction accuracy and efficiency are improved by about 20% compared with the LSTM prediction model.