| Summary: | Streaming video comprises the most of today’s Internet traffic, and it’s pre-
dicted to increase more. Today millions of users are watching video over
the Internet; video sharing sites are getting more than billion hits per day.
To serve this massive user base has always been a challenging job. Over
the period of time a number of approaches have been proposed, mainly in
two categories - client server and peer to peer based streaming. Despite the
potential scalability benefits of peer to peer systems, most popular video
sharing sites today are using client server model, leveraging the caching
benefits of Content Delivery Networks. In such scenarios, video files are
replicated among a group of edge servers, clients’ requests are directed to
an edge server instead of serving by the original video source server. The
main bottle neck to this approach is that each server has a capacity limit
beyond which it cannot serve properly.
Instead of traditional file based streaming approach, in this thesis we pro-
pose to use distributed data structure as the underlying storage for streaming
video. We developed a distributed B-tree, and stored video files in the B-
tree which runs over a cluster of computers and served from there. We show
that system throughput increases almost linearly when more computers are
added to the system.
|
|---|
