Analysis of an interactive video architecture

Citation

Mouchtaris, Petros N. (1993) Analysis of an interactive video architecture. Dissertation (Ph.D.), California Institute of Technology. doi:10.7907/A9Z4-N267. https://resolver.caltech.edu/CaltechETD:etd-02022005-162907

Abstract

A new residential application for interactive video is proposed. There is a service provider that prepares and distributes daily news programs customized to subscriber interest. The provider assembles the programs from short news clips and uses a profile data base of subscribers for selecting the appropriate clips. The time of viewing the program can be selected by the customers in near-real-time. We model this service and propose a network architecture that can support it. There is a main node that contains most of the storage and sourcing facilities, and an intermediate node to which all customers are connected. Multicasting is used as much as possible for reducing the traffic load on the network. In addition to that, popular material is stored in the intermediate node which is closer to the customers, which further decreases the traffic load.

Our main concern is the time that a customer has to wait until he starts getting his program. This time is a function of the capacity of the link that connects the main node to the intermediate node, the so-called main link. The case that the main link can only transport a single video connection is considered first. We propose a recurrent algorithm that calculates the probabilities of the states and uses them for evaluating the expected wait, and prove that there is a very simple relationship between the expected wait and the probabilities of the states. A simplified analysis that directly computes the expected wait is proposed next. This approach is computationally more efficient but does not give us any information about the probabilities of the states. For the general case that the main link can transport more than one video connection, we generalize the recurrent algorithm that calculates the probabilities of the states and the simple relationship between the expected wait and the probabilities of the states. For the cases that the complexity of our algorithm is too large, we propose and evaluate three approximate techniques for estimating the expected wait. In the first technique we use the results for the case that a main link can only transport a single connection for estimating the results for the general case. In the second technique we use the idea of rescaling time. In the third, motivated by the fluid-flow theory, we solve a deterministic problem and use the results of that problem for estimating the expected wait for the problem we are interested in. We show that these approximate techniques compare well with simulations. Thus, we can now decide what the capacity of the main link should be so that our system has the desired performance, and we can do that even if the number of customers is very large.

Thesis Files