Title:

Optimizing content dissemination in federated online social networks

Description:

In recent years, federated online social networks are gaining popularity. However, along with the increase of user numbers, the content dissemination problems of federated online social networks have become more and more obvious. In this thesis, we focus on optimizing the content dissemination in federated online social networks by reducing the whole network traffic and the traffic load for individual small servers.
This work proposed Mufesa (Multilevel Feedback Queue Scheduling Algorithm), a solution that can classify different types of contents and then transmit them using various strategies. Specifically, Mufesa classifies the contents into four categories according to their timeliness requirements and their size and processes these contents in four job queues with different priorities. Each queue has its own scheduling algorithm. Instant messages and Voice/Video-Calls are those contents that should be sent immediately and will be put into the queues with the highest priority. There are two queues with the highest priority. The one using the Roundrobin (RR) Scheduling is designed for instant message and its subtypes. The one without time slice is for Voice/Video-Calls. Posts, such as images, videos, texts and so on, are those contents that have relatively lower priority than instant messages and will come to the queue with the middle priority. Size-based transmission approach will be used in this queue. Other OSN features such as like, follow, report and so on will come to the queue with the low priority, but can be piggybacked by contents in other queues. Besides, the queues with the middle and low priority have different timers.
We have implemented Mufesa in a federated online social network called PeerTube to provide a proof of concept. Our evaluation using real datasets shows that Mufesa can efficiently reduce the amount of traffic in the whole network by 31.05%, can minimize the traffic load for small home servers and is also able to reduce the number of requests by 13.65%.
Keywords: Federated online social networks, content dissemination, content type classification, timer approach, piggybacking, PeerTube

Supervisor: Sebastian Göndör [1]

Type:  Master Thesis

Duration: 6 months