Design And Implementation Of Fault-Aware Media Mixing Server Cluster

With the continuous development of mobile internet technology,realtime audio and video communication has gradually been valued by the public in recent years.The architecture pattern of using a Multipoint Control Unit(MCU)as a central server is a mainstream architecture pattern in multi-party audio and video conference scenarios.In this architecture,MCU provides audio and video mixing service for conferences.The media processing capacity of a single media mixing server is limited,so this paper designs a media mixing server cluster that can horizontally expand the media processing capacity of video conferencing system.In order to proactively detect faults in servers and improve the high availability of the cluster,this paper studies the problem of server fault detection,and uses a multivariate time series anomaly detection method based on the characteristics of cloud server status to achieve server fault detection.The main work of this paper is as follows:(1)An unsupervised multivariate time series anomaly detection network based on the autoencoder structure is used.(2)Clustering algorithm is used for data preprocessing,and a normal sample recall strategy is proposed to adjust the clustering algorithm results,which can reduce the number of abnormal samples in the test dataset and improve the robustness of the model.(3)Using graph attention networks in autoencoder networks to model the correlations between time-series data,and extracting time-series features in both the time domain and frequency domain.The above approach can extract complex spatiotemporal dependencies in multivariate time series.(4)A media mixing server cluster is designed and implemented,which can provide stable media mixing services.The usability and reliability of the anomaly detection algorithm in practical scenarios are verified on the system,and functional testing and performance testing are performed on the system.