| As a lightweight network virtualization technology,container networks have achieved a wide range of applications in digital twin networks,metaverse,network emulation,and other fields.Therefore,the measurement of traffic running in container networks is of great significance in network troubleshooting,network performance analysis,and network traffic management.Traffic measurement in container networks can be divided into four stages: collection,analysis,storage,and visualization.Existing traffic measurement schemes do not fully exploit the characteristics of container networks,and have the disadvantages of high overhead and difficult deployment when collecting and analyzing;in addition,in the face of the demand for storage and visualization of traffic data,most traditional measurement schemes rely on manual organization and visualization of traffic data,which takes a lot of efforts of users.To address the above problems,this thesis proposes solutions for all four stages of traffic measurement for container networks.For the collection and analysis stage,based on the observation that container networks usually share the same server environment,this thesis innovatively proposes a fine-grained,lightweight,network-wide and easy-to-deploy traffic collection and analysis scheme by overcoming the design conflicts of existing schemes.The scheme 1)is implemented in the kernel space based on eBPF technology,thus avoiding the huge CPU overhead of copying packets from the kernel space to the user space when performing finegrained measurements.2)allocating a corresponding storage space in memory for each packet to store the metadata which is necessary to calculate traffic metrics,greatly reducing the virtual network bandwidth overhead caused by carrying metadata inside packets in existing schemes.3)rewriting some existing bits in the IP packet header to achieve network-wide identification of each packet without increasing the packet size,thus achieving easy deployment without modifying the configuration of the NIC.Experiments show that the scheme can measure fine-grained,network-wide traffic metrics with ease of deployment and low overhead.For the storage and visualization stage,this thesis designs and implements a lowoverhead,user-friendly traffic storage and visualization scheme to address the problems of high read/write overhead and manual organization and visualization of measurement data in existing schemes.The scheme consists of a traffic metrics storage module and a traffic metrics visualization module.The traffic metrics storage module 1)is developed based on a time-series database,and the write performance is optimized by using multithreading and batch writing techniques.2)design a data aggregation strategy to reduce the pressure when reading data.3)implement a automatic data cleaning mechanism to reduce the storage space occupation.4)provide various APIs for users to support the efficient operation of the storage module.The traffic metrics visualization module provides two interfaces,data overview visualization and data detail visualization,and can generate various types of charts for traffic metrics to meet the needs of data quick overview and complex visualization,which improves user-friendliness.Experimental results show that the traffic storage and visualization scheme has low read and write overhead,and users can easily visualize the traffic metrics through the scheme to get the status information of the traffic in the container network quickly. |
