An intelligent virtual network path planning method for large-scale container clusters

Abstract

The invention relates to an intelligent virtual network path planning method for large-scale container clusters. After extracting the interface information, the entire network information is sensed, and the topology information is reconstructed; traffic statistics are performed on switches; After the statistics, the corresponding flow table information is automatically extracted to obtain two types of features including the number of inbound bytes of port traffic and the number of outbound bytes of port traffic, and two types of self-built features are constructed, namely, the inflow rate of port traffic and the outflow rate of port traffic. Normalize the input features, and obtain the traffic prediction model after training; use the traffic prediction model to predict the port traffic of the switch in a fixed time period in the future according to different tasks, and compare the prediction result with the threshold value to the corresponding switch Perform sleep / start operation; carry out adaptive planning according to the characteristics of the task; execute the action strategy to be taken.

Description

technical field [0001] The invention relates to container network technology, in particular to an intelligent virtual network path planning method for large-scale container clusters. Background technique [0002] Due to the poor isolation of network resources in the container network itself, when the scale of the container becomes large, when the network is isolated, using traditional routing strategies for path planning will result in a large-scale routing table, low efficiency, and the inability to perform intelligent tasks based on task characteristics. Network path planning and selection. When carrying out network communication between containers across hosts, the traditional method is mainly through integration tools or components. One representative method is to store network configuration information including network segments and subnet numbers in etcd, a distributed key-value pair storage system, and use etcd to Store the corresponding relationship between differen...

AI Technical Summary

Problems solved by technology

[0003] Although the existing solutions can solve the problem of cross-host communication of containers, in terms of isolation of network resources, method 1 is used to find the correspondence between the subnet and IP stored in etcd. Although the cross-host communication of containers can be realized, there is no network Isolation capability; while using a virtual network to connect and deploy containers located on multiple hosts in the form of virtual bridges and routers, it can have network isolation capabilities while realizing cross-host communication of containers

However, neither of these two methods can make flexible and rapid adjustments to business needs according to the characteristics of the task, especially when the scale of the container is relatively large, the corresponding relationship between node storage or the scale of the virtual routing table is large, and when the container communicates , because the table lookup will bring a large computing overhead, which leads to a decrease in communication efficiency. At the same time, when large-scale container communication occurs, the traffic distribution is unbalanced and the lack of reasonable scheduling will make most switches in the network even under low load conditions. It will continue to work, causing unnecessary energy loss

The existing network node data plane and control plane are tightly coupled. Network virtualization has problems such as heavy deployment workload and difficulty in making flexible and fast adjustments to business requirements. The isolation of network resources cannot meet the needs of adaptive tasks.

Images