Characterization and Data Analysis of Cloud Gaming Platforms

Gaming has evolved as one of the most used and profitable online business worldwide. The players usually access their games through their PCs, laptops, phones or tablets; however, recently some companies like Google or NVIDIA have launched a new model called cloud or browser gaming. The idea is to run the game in the company servers and stream the content of the game to the user. Little to no information has been published about how they are implementing this new service, so this thesis is aimed at doing a traffic characterization of cloud gaming and with the help of machine learning algorithms, determine in which state was the game at any second so in the future some QoE mechanism can be applied to improve the performance. The thesis has been organized as follows, firstly, the focus was on Data collection, the more data the better to perform a correct characterization of the service, also it is very useful for building the datasets later on for the machine learning estimations. The next step was developing python software that allowed the extraction of the information from the Wireshark captures and the WebRTC logs, this information was used to build the dataset, later on a target column was added manually using the video recordings of the games for distinguishing between phases inside the games. Finally, classification machine learning algorithms were trained to predict if the user was actively playing or not (pauses, menus or loading up phase).