BOSTON UNIVERSITY GRADUATE SCHOOL OF ARTS AND SCIENCES Dissertation SCALABILITY OF MULTICAST-BASED STREAMING DELIVERY MECHANISMS ON THE INTERNET

This thesis examines the scalability of multicast-based streaming delivery through analytical and empirical evaluation methodologies. Scalability is assessed with respect to two metrics: server bandwidth and network cost. The first metric quantifies the amount of server resources needed to serve a large number of concurrent clients. The second metric quantifies the amount of network resources needed to serve these clients over the Internet. Scalability along these metrics is evaluated subject to two types of characteristic properties: access patterns and Internet topology. Access patterns assumed in this thesis allow clients to be synchronous or asynchronous, and allow them to access content sequentially or randomly. Internet topologies considered in this thesis exhibit power-law vertex degree distributions and small-world behavior. The findings of this thesis show that the server bandwidth requirement of multicast-based delivery techniques–such as stream merging and periodic broadcasting–largely depends on client access patterns. In particular, for asynchronous clients, if access is sequential, then the lower bound on server bandwidth grows logarithmically with the request arrival rate, but if client access is random, then the lower bound grows as the square root of the request arrival rate. The thesis also shows that the network cost of multicast-based streaming delivery depends mainly on the Internet topological properties. Specifically, both Internet power-law vertex degree distribution and small-world behavior affect the scaling behavior of IP multicast and of end system multicast mechanisms.