注意力机制LSTM虚拟机能耗建模方法

Energy consumption modeling method of attention mechanism LSTM virtual machine

云计算平台因海量资源池带来了巨大的能耗开销。以虚拟机为粒度,进行灰色关联度分析,采用Hypervisor技术监控虚拟机运行状态参数,引入注意力机制进行LSTM虚拟机能耗建模,模型激活函数采用LeakRelu函数。实验数据呈现能耗模型的实时功率平均误差为5.6%。实验模型对比LSTM、MLP、SVM及K近邻算法,选用WordCount与Sort任务进行虚拟机能耗模型测评,实验结果表明,能耗建模质量优于LSTM、MLP、SVM及K近邻算法。

The cloud computing platform causes huge energy consumption due to its massive resource pool. Virtual machine(VM) was used as the granularity, grey relation analysis was implemented, Hypervisor technology was selected to monitor VM operating state parameters, and an attention mechanism was introduced to model energy consumption of LSTM virtual machines. The activation function used the LeakRelu function. Experimental data show that the average real-time power error of the energy consumption model is 5.6%. Furthermore, the experiment used WordCount and Sort tasks to compare the quality of VM energy consumption modeling for LSTM, MLP, SVM and KNN algorithms. The results show that the quality of energy consumption modeling is better than that of LSTM, MLP, SVM, and KNN algorithms.