| dc.description.abstract | With the advent of the 5G era, two key technologies, Software-Defined Networking
(SDN) and Machine Learning (ML), are becoming more and more mature. Through
SDN, network operators can allocate network functions and adjust topology flexibility
more effectively and flexibly. On the other hand, by classifying different learning styles
and learning tasks by ML, it can be applied in many different fields, and the data can be
analyzed and corresponding strategies deployed in a relatively short time. However, at
the same time, because of the 5G big data era, new problems and challenges have emerged
in the two fields. In terms of SDN, taking the most common protocol OpenFlow as an
example, in order to adapt different packets formats, the header has become more and
more complex, and the deployment has become more and more complex; In the aspect of
ML, the convergence time begins to be lengthened because of the learning large amount
of data, and the problem of private data has not been effectively solved, so the model
performance needs further improvement.
With the invention of P4 switch, the original Bottom-Up switch design has been
changed to Top-Down, which has the advantage that the protocol of the switch is no
longer bound to the hardware, thus making more effective and flexible changes to the
action rules of packets. In recent years, the emerging Federated Learning (FL) framework
can not only keep the privacy of the terminal effectively, but also carry out effective model
training, overcoming the problems of the previous centralized and decentralized learning
framework. This study will also make use of the advantages of P4 and FL mentioned
above to design an environment to optimize the packet transmission of the Internet of
Things (IoT). Through the experimental results, it can be found that compared with the
control group, the environment set in this paper is the best, so it can be proved that these
two technologies can effectively improve the performance of packet transmission of the
IoT. | en_US |