Efficient Streaming for Delay-tolerant Multimedia Applications

Delay-tolerant multimedia applications, where clients specify the time when playout must start, fit the profile of many emerging applications, including distance education and corporate training. Such applications typically depend on a Content Delivery Network (CDN), where the Content Service Provider (CSP) disseminates multimedia content to geographically dispersed clients through a distribution network from servers located on the CDN. In these applications, a client Ci connects to the source S at time t0 and requests for contents with base encoding rate Γ and playout duration T ; Ci specifies the time it is willing to wait for the playout to start, its delay tolerance δi; Playout at Ci starts at (t0+δi) and ends at ((t0 + δi) + T ). Note that t0 may also be the start time of a scheduled streaming session. This thesis deals with the issue of maximizing the quality of the multimedia contents delivered at the clients even when link bandwidths are constrained in the path from the source to the clients. To achieve this, we develop a suite of algorithms that can exploit clients’ delay tolerance considering the following parameters: (i) Service type: whether contents are streamed according to a schedule or occur on-demand and (ii) Bandwidth: whether link capacity is static or variable, using appropriate resources at the nodes – Transcoders, Layer encoders, or Streaming servers with transcoding or layering functionality. In particular, we consider the following three cases: (i) Scheduled streaming when bandwidths are static, (ii) Scheduled streaming when bandwidths are varying and predicted, and (iii) On-demand streaming when bandwidths are static. The algorithms developed are a result of formulating the objectives in an optimization framework or by an analysis of properties of the network topology and client requirements. Furthermore, where an optimal algorithm is computationally expensive, we develop algorithms based on heuristics as practical alternatives. The approaches are validated through extensive simulations, using topologies derived from real-world scenarios. The algorithms developed in this thesis would help a CSP serving clients in a subscription based network in: (i) improving the quality of reception at the clients by leveraging their delay tolerance values, (ii) estimating the resources required to provide the best delivered rates to clients, and (iii) determining placement of resources to maximize the delivered rates at clients. Thus, using the analysis presented and algorithms developed in this thesis, a CSP can deploy resources in order to ensure effective quality of service to clients and efficient resource utilization by leveraging clients’ delay tolerance.