The last three years have seen numerous proposals aimed at reducing the cost of
video-on-demand services. Most of these proposals have focussed on lowering the
bandwidth requirements of these services without increasing customer waiting times.
Few of these proposals have addressed the problem of distributing moderately popular
videos or videos whose popularity fluctuates over time.
We propose in this thesis a new distribution protocol that addresses both problems. Our
Dynamic Reactive Broadcasting protocol combines reactive and proactive approaches
in order to distribute moderately popular videos in the most efficient fashion. In
addition, it is the first video distribution technique that adapts itself to changes in the
request arrival rate by changing its distribution strategy.
Our simulation study indicates that our new protocol performs as well as or better than
any other distribution protocol as long as the request arrival rates remains under 380
requests per hour. Dynamic Reactive Broadcasting should thus be the method of choice
for distributing all videos but these that will remain in great demand over several hours.