A Study Of Software--defined Network Measurement Based On Federated Learning

In recent years,Software-Defined Networking has developed rapidly,and network measurement technologies based on Software-Defined Networking has also begun to at-tract more attention,a series of research results have emerged.However,the problem is that a large number of algorithms have not been well applied.Because there are some defects in flexibility and scalability.In order to solve this problem,this paper in-troduces Federated Learning into the Software-Defined Networking measurement task,and designs a network measurement framework FLSketch based on Federated Learning nd multi-dimensional time series sketch.The framework combines Federated Learning with traditional sketch network measurement methods,and it has better flexibility and scalability than traditional methods.The work of this paper mainly includes three as-pects:1.A network measurement framework FLSketch based on Federated Learning and multi-dimensional time series sketch is designed,and a multi-dimensional time series sketch data structure suitable for the framework is proposed.The implementation methods of the FLSketch framework in the estimation of the number of data packets,the total size of the data packets,the average size of the data packets,the average delay estimation,the packet loss rate estimation and the DDo S attack detection tasks are given.2.In the traditional DDo S attack detection,the sketch algorithm has poor detec-tion sensitivity and uncertain space cost,and the data stream-based machine learning attack detection is computationally intensive.This paper designs a two-stage detection algorithm based on the FLSketch framework.Some attack detection experiments were carried out on the sketch detection granularity and the data flow granularity.The accu-racy of the experiment reached a level similar to the traditional packet-based machine learning algorithm.But it greatly reduces the calculation amount and improves the ef-ficiency of detection.The experimental results show that the DDo S attack detection algorithm based on the FLSketch framework has a good performance.3.The FLSketch framework is divided into modules,and the prototype of the FLSketch measurement framework is realized.The implementation details of the multi-dimensional time series sketch algorithm in the framework are introduced in detail,the measurement implementation of the FLSketch framework for common network mea-surement tasks is introduced.The pseudo code of the Federated Averaging algorithm and the data flow filtering algorithm in this article are given.