Cognitive networking techniques on content distribution networks
Tutor / director / evaluatorGarcía López, Pedro
Document typeMaster thesis
Rights accessOpen Access
First we want to design a strategy based on Artificial Intelligence (AI) techniques with the aim of increasing peers download performance. Some AI algorithms can find patterns in the information available to a peer locally, and use it to predict values that cannot be calculated by means of mathematical formulas. An important aspect of these techniques is that can be trained in order to improve its interpretation of the local available information. With this process they can make more accurate predictions and perform better results. We will use this prediction system to increase our knowledge about the swarm and the peers who are part of it. This global knowledge increase can be used to optimize the algorithms of BitTorrent and can represent a great improvement in peers download capacity. Our second challenge is to create a reduced group of peers (Crowd) that focus their efforts on improving the condition of the swarm through collaborative techniques. The basic idea of this approach is to organize a group of peers to act as a single node and focus them on getting all pieces of the content they are interested in. This involves avoiding, as far as possible, to download pieces that any of the members already have. The main goal of this technique consists of reaching as quickly as possible a copy of the content distributed between all members of the Crowd. Getting a distributed copy of the content is expected to increase the availability of parts and reduce dependence on the seeds (users who have the complete content), which would represent a great benefit for the whole swarm. Another aspect that we want to investigate is the use of a priority system among members of the Crowd. We consider that in certain situations to prioritize the Crowd peers at expense of regular peers can result in a significant increase of the download ratio.