Research On Resource Quantification And Topology Mapping For Containerized Network Emulation Platform

With the booming development of the Internet,network structures and mechanisms are getting richer and richer,which makes network modeling and protocol emulation more and more complex.To meet the growing demand for network experiments,containerized network emulation platforms have emerged.Such platforms need to allocate appropriate physical resources to container nodes and make decisions on node placement when deploying experiments to ensure that experiments can be conducted properly.If the resources are not properly allocated or mapped,the fidelity of the experiments will be affected.Therefore,reasonable resource allocation and efficient topology mapping are the cornerstones for successful completion of experiments in the platform.However,the existing resource allocation work for containerized network emulation platform usually adopts only static allocation,which lacks flexibility,and similar topology mapping algorithms do not fully consider physical resource constraints.To address the problem of improper resource allocation or mapping for network emulation platforms,this thesis focuses on the characteristics of container networks,proposes a resource quantification and topology mapping scheme for containerized network emulation platforms,and verifies the reasonableness and efficiency of the scheme.The work is as follows:To address the resource quantification problem,this thesis focuses on the resource overhead characteristics of container network elements and proposes a semi-automated resource quantification scheme by integrating the ideas of static prediction and dynamic adjustment.The scheme accomplishes efficient and reliable quantification of container network resources from two aspects:resource estimation and resource monitoring and adjustment.On the one hand,the resource estimation scheme takes into account the problem that it is difficult to directly estimate the resource demand of the application part because of the great variety of applications in containers,and uses a static configuration to complete the pre-allocation of resources for the application part;on the other hand,considering that the characteristics of sending and receiving packets are basically the same for any type of container,a longest path-based communication requirement quantification algorithm is designed to estimate the resource overhead due to sending and receiving packets for the resource requirement of the communication part.Further,in case of inaccurate resource estimation,a resource monitoring and adjustment scheme is proposed to obtain the resource usage in real time and make dynamic resource adjustment in time.Finally,the effectiveness of the resource quantification scheme is verified by specific use cases.To address the topology mapping problem,this thesis proposes an efficient topology mapping scheme considering the characteristics of the containerized network emulation platform in detail.First,a container network mapping model and the basic principles of mapping are proposed,and a mapping algorithm based on multi-level graph partitioning is designed.The algorithm can actively sense the resource margin of physical clusters and adjust the allocation policy of graph partitioning to improve the resource utilization of physical clusters while ensuring resource constraints.Further,a reasonable isolation optimization scheme is proposed to address the problem of poor isolation of container networks,so as to effectively reduce the performance interference among containers.The experimental results show that the algorithm performs well in terms of deployment performance compared to the comparison algorithm,and is able to improve the deployment success rate by up to 34.41% while reducing the server requirements by 10% to 600%.In terms of experimental capacity,the algorithm is able to support a larger number of topologies with an increase of up to 350% and achieve a significant increase of 166% in the number of nodes.