| Summary: | 碩士 === 國立臺灣大學 === 電機工程學研究所 === 104 === With the rising popularity of SDN (Software-defined Networking) and machine learning, we are motivated to apply these two things to peer-to-peer (P2P) network to see what it can do for P2P network.
Considering the large-scale deployment of SDN nowadays is still a big problem, we construct our environment by the combination of SDN network and traditional network rather than using SDN network for whole environment only.
This thesis proposes an incentive policy to reinforce the existing incentive policy in BitTorrent system and the goal of this thesis is to decrease the traffic of bad users as much as possible. We emulate the network in Mininet and several BitTorrent users with different user behavior. The data center collects information comes from switches, hosts, and the tracker and use machine learning model to classify the type of user behavior in each period. The data center also derives a score for each user, and give punishments or rewards to them according to their score. The punishments and rewards are presented in the form of quality of service (QoS), and the task of adjusting QoS is achieved with the help of SDN and Ryu-QoS.
There are 65 hosts distributed in our experimental environment. Almost all of them are all distributed in the traditional network, but one of them is distributed outside the network we emulated to provide the source of data. We can see the result of our experiments from the curve of average download speed of all bad users, which exactly decrease after our punishments.
|
|---|
