Research On Virtual Network Performance Optimization Under Cloud-edge Collaboration

In recent years,with the development of 5G and related transmission technologies,the transmission speed has been greatly improved,and the "Internet of Everything" scenario is no longer a distant dream.In order to make better use of the advantages of 5G transmission speed,edge computing relying on cloud computing has also begun to show its head.But at present,the most common starlingx edge computing platform has the disadvantage that there is no direct communication between different virtual architectures,and the communication performance between the same virtual architectures is low,which shows that there is still a lot of room for development in edge computing technology..In order to solve the above-mentioned problems,this paper proposes a complete communication optimization system,which performs communication on the edge computing platform starlingx from three angles of the same physical host inside the edge cloud,different physical hosts between cloud edges,and physical host deployment solutions.Optimized so that it can communicate directly and efficiently even in various virtual architectures.For the communication optimization of the same physical host in the edge cloud,this paper proposes three solutions: extended calico network solution,kernel mapping solution,and shared memory solution.After comprehensive consideration,the system chose a shared memory solution,which is to open up a memory in the same physical host for all virtual machines and containers to share,and achieve the purpose of direct communication by operating the shared memory.For the optimization of different physical host communication between the cloud sides,this paper proposes a simple network solution and an RDMA network solution.After a comprehensive discussion,the system selects the RDMA network solution,that is,directly writes the data to be sent through the RDMA technology.The memory of different physical hosts reduces the loss caused by data parsing through the network.At the same time,in order to combine the RDMA network solution and the shared memory solution,the system also proposes the IPC server solution,which is responsible for pushing the shared memory in the shared memory solution to the RDMA network solution,and can automatically send the data in the shared memory.Different physical hosts can also obtain the data to be transmitted in time.On different cloud-side platforms,how to reasonably deploy virtual nodes that have a communication relationship with each other has also become an urgent problem to be solved.In this paper,on the basis of proving that this deployment problem is an NP problem,the two common deployment requirements for communication-first deployment and performance-first deployment are related heuristic algorithms,which provide theoretical basis and practical support for thistype of problem.After the system is completed,this article gives a test of the various functions of the system and evaluates the system performance.In the same physical host,the communication optimization system reduces the delay of 5% compared to kafka transmission;in different physical hosts,the communication optimization system reduces the delay of 10% compared to kafka transmission;if a deployment module is adopted to perform virtual node pre-processing,it reduces the delay by 5% compared to the random deployment scheme.This proves that the communication optimization system does achieve the purpose of optimizing communication.