Scalable Distributed Network Protocol Testing And Verification System

In recent years,with the continuous expansion of cloud computing,and the gradual popularity of programmable data planes,the network protocols in data centers are undergoing rapid changes.In order to ensure the stability and reli-ability of new protocols,a specific group of engineering tests and verifications are needed before their deployment.Due to the high cost of using real testbeds,an emulation environment is usually chosen to test and verify the function and performance of the network protocol.In this process,three problems will be encountered:First,because of the considerable scale of the networks,as well as the limitation on the performance which traditional emulation technologies are able to achieve,carrying out a real traffic emulation of its entire network is not always a practical solution.Second,since most of these technologies try to focus on emulation,to some extent,they fail to meet the needs of network proto-col testing and the verification of it.Finally,the network telemetry technology in the distributed emulation environment needs to be further improved.Therefore,this thesis proposes a design scheme of distributed network pro-tocol emulation and testing system for the possible improvement.In the first part,a solution based on container and tunnel technology will be introduced.In such a solution,nodes and links in an emulated network will run on multiple hosts synchronously,in order to solve the scalable issue during emulation of a large-scale network.In the second part,as an application of eBPF technology,a new out-band network telemetry method,which can efficiently obtain running data by the filter and extract packets from network nodes for debugging and analysis purposes,will be proved to be a successful technique under distributed emulation scenario.In the end,this thesis proposes a protocol test method that is based on test cases and discusses some additional pros which can be gained by implementing such a method in a test-driven manner.Essentially,the functionality and performance were tested by simulating the procedure of adding a new protocol to a data center' s network followed by some properly designed tests.The data obtained from these tests indicate that the model can full fill at least three requirements,including distributed emula-tion,real-time telemetry,and the verification of protocols which are running in large-scale networks.Performance test proved that it is sufficient to emulate a network,which has about 640 nodes and 1280 links,on a single host whose resources are set to a dual-core CPU and a 4GB memory are available to use.However,the performance still can be dramatically increased if there are mul-tiple nodes present in the emulation networks.