Large-scale Virtual Network Orchestration Management Research For Containerized Network Emulation Platform

Network emulation platform has been widely adopted in the study of network problems,However,the existing containerized virtual network emulation platform still has major shortcomings.On the one hand,current state-of-art network emulation platforms do not scale beyond a single machine while others that support large-scale network emulation only support for specific network scenarios.On the other hand,the existing virtual network emulation platform is not reliable enough,and lacks the ability to monitor and recover the container network and platform service processes,which will incur high experiment time costs and troubleshooting costs in large-scale experiment scenarios.Therefore,this thesis focuses on the shortcomings and deficiencies of the current virtual network emulation platform based on container virtualization technology,and designs orchestration management methods for large-scale virtual networks and high-availability methods for large-scale virtual network services from the perspective of supporting large-scale virtual network experiments,and the main research points and contents are as follows:First,in view of the limited experimental scale supported by the existing virtual network simulation platform,this thesis designs a large-scale virtual network orchestration management system.A cross-host virtual network networking scheme is given,which supports the orchestration and deployment of virtual networks on different hosts,so that the scale of virtual networks can be easily expanded.At the same time,the thesis desgins a virtual network data model suitable for distributed deployment across multiple hosts to support rapid extraction and modification of part of the topology state information of the virtual network deployed on a specific host,so as to reduce the management cost of maintaining large-scale virtual network topology state information.Furthermore,in order to reduce long deployment time of large-scale virtual network,this thesis proposes a concurrency deployment framework for orchestration tasks based on coroutines,which orchestrates tasks of multiple virtual networks on the master node,and process multiple virtual nodes and link deployment tasks are processed in concurrency to speed up the orchestration and deployment of large-scale virtual networks.Secondly,aiming at the lack of high availability of the large-scale virtual network emulation platform,this thesis analyzes the control framework of the management plane in the above-mentioned virtual network orchestration management system,the risk that the orchestration service is unavailable due to a single point of failure in the data storage module,and The virtual network of the data plane has the risk of being inconsistent with the experimental network expected by experimental users due to container failure.Aiming at the high-availability defect of the management plane,this thesis designs and implements a multi-decision monitoring process that prevents misjudgment to monitor the control framework service process,and combines NGINX and Keepalived to provide high-availability services for the control framework,and introduces the Redis sentinel mode to solve data storage For the problem of high availability,an failover solution is realized; for the high availability problem of the virtual network of the underlying emulation plane,a virtual network monitoring and automatic fault recovery solution based on polling check is designed and implemented.Finally,this thesis combines the actual use case scenarios to verify the function and performance of the large-scale virtual network orchestration management system.The verification results show that the virtual network orchestration management system can well support large-scale virtual network experiment scenarios and can speed up the virtual network orchestration process with lower CPU resource overhead.Compared with multi-process and multi-thread concurrency technology,it can achieve almost the same concurrent orchestration effect with lower CPU overhead.Compared with multi-process,the maximum possible reduced by 89.94%,compared with thread,the maximum can be reduced by 68.04%.At the same time,the system has high availability at the process level.