基于注意力时空卷积和A2C的虚拟机主动容错优先迁移决策

Active fault-tolerant priority migration decision model for virtual machinesbased on attentional spatio-temporal convolution and A2C

针对边缘云环境的自动化和分布式特性、高度不可靠性及易变的工作负载问题,提出基于注意力时空卷积和A2C的虚拟机主动容错优先迁移决策模型AST-A2C。首先,采用带有注意力机制的长短期记忆网络(LSTM)提取各主机的时序特征,根据时序特征和多主机间的交互信息构建图网络,再利用图注意力网络(GAT)提取网络中不同主机间的关联信息,将其用于主机的故障信息编码。其次,设计可动态建立模型并不断生成改进决策的A2C模块,联合故障编码信息和调度决策信息进行优先迁移决策。最后,构建满足不同用户QoS要求和应用程序设置的自适应损失函数来优化调度决策。实验结果表明,该模型在故障检测、能源消耗、时延敏感性等方面优于最先进的基线,是提高边缘云计算可靠性的理想选择。

This paper proposed an active fault-tolerant priority migration decision model for virtual machines based on attentional spatio-temporal convolution and A2C(AST-A2C).This method improved reliability and adapted to volatile workloads in resource-constrained edge cloud computing environments.Firstly,it used a long short-term memory network(LSTM)with an attention mechanism to extract the temporal features in each host.It combined temporal features and interaction information between multiple hosts to build graph networks.Then it used graph attention network(GAT)to extract association information between different hosts in the network,and the association information would be used to encode the fault information of hosts.Next,it designed the A2C module to dynamically build models and continuously generate improved decisions,this module could combine fault code information and scheduling decision information for priority migration decision.It constructed adaptive loss functions to optimize scheduling decisions to meet different user QoS requirements and application settings.Experimental results show that the proposed model outperforms the state-of-the-art baseline in terms of fault detection,energy consumption,and delay sensitivity.It is ideal for improving the reliability of edge cloud computing.