In order to solve the problem that the centralized cloud computing technology can not achieve the computing bandwidth of a large number of edge data and can not guarantee the privacy and real-time of the application, this paper systematically analyzes the time difference of the edge container cloud load under the condition of multi-cluster and the edge application with different latency sensitivity requirements, and provides a multi-cluster edge cloud architecture managed by a master-slave model. The resource allocation related to the delay-sensitive application is studied in depth, and the proposed related research can be effectively realized by comparing the existing responsive strategies; for the delay-sensitive application, the load is extended in advance at the upper edge, or the lag is reduced at the lower edge of the load, so as to effectively meet the needs of the application quality. The results show that the distributed edge computation mode can improve the related computing power in the sink cloud around the actual application, reduce the computing load of the cloud center itself, and reduce the bandwidth pressure of the core backbone network.