| Resumo: | One of the goals of an operator is to improve the Quality of Experience (QoE) of a client in networks where Over-the-top (OTT) content is being delivered. The appearance of services like YouTube, Netflix or Twitch, where in the first case it contains more than 300 hours of video per minute in the platform, brings issues to the managed data networks that already exist, as well as challenges to fix them. Video traffic corresponds to 75% of the whole transmitted data on the Internet. This way, not only the Internet did become the ’de facto’ video transmission path, but also the general data traffic continues to exponentially increase, due to the desire to consume more content. This thesis presents two model proposals and architecture that aim to improve the users’ quality of experience, by predicting the amount of video in advance liable of being prefetched, as a way to optimize the delivery efficiency where the quality of service cannot be guaranteed. The prefetch is done in the clients’ closest cache server. For that, an Analytic Hierarchy Process (AHP) is used, where through a subjective method of attribute comparison, and from the application of a weighted function on the measured quality of service metrics, the amount of prefetch is achieved. Besides this method, artificial intelligence techniques are also taken into account. With neural networks, there is an attempt of selflearning with the behavior of OTT networks with more than 14.000 hours of video consumption under different quality conditions, to try to estimate the experience felt and maximize it, without the normal service delivery degradation. At last, both methods are evaluated and a proof of concept is made with users in a high speed train.
|
|---|