Asynchronous multimedia messaging

This thesis is an engineering study of the problem of providing asynchronous multimedia messaging within the context of the Internet. In part one, we consider the problem of adding continuous media (audio, video, animation, slide show, etc.) to Internet email, and we present a voice message board prototyped as a Web service. Central to asynchronous multimedia messaging systems is the streaming delivery of stored data. In part two, we discuss optimal transmission policies for streaming scalable asynchronous multimedia data. In part one, we demonstrate the inadequacy of the current email infrastructure for supporting continuous media data. We solve the problems with a sender-side storage and streaming delivery mechanism that can be incrementally implemented in the Internet by adding new functionality only to sender systems. This new storage and delivery model introduces several new problems, including reduced media quality, message deletion, replying, and forwarding. We propose solutions to these problems. Also in part one, we present the prototype voice message board aconf, and discuss the design constraints and alternatives. In part two, we formulate the problem of streaming stored scalable multimedia data as a bandwidth and client memory constrained optimization. We use the term multimedia presentation to mean any stored multimedia data, including message data. Multimedia presentations are collections of multimedia data units with rendering intervals relative to the start of the presentation. In order to reduce the delay between user request for playout and the start of playout, the application streams the data to the client, that is, it begins playout at the client after commencing transmission of the data. In order to reduce the start up delay as much as possible in the context of limited bandwidth, the application scales down the data representation of the media so that the component units will arrive before their rendering intervals start. Scaling also becomes necessary when client memory capacity is relatively small, limiting the amount of data that can be pre-fetched before their playout intervals commence. When client memory is an issue, the application must determine both the magnitude of the scaling, and the rate of transmission relative to the presentation timeline. Scaling is also necessary when implementing bandwidth adaptation schemes in the Internet. Internet bandwidth is not generally predictable, and so applications must scale down the size of the data representation during times when bandwidth decreases, and scale up when bandwidth increases. We formulate the refined max-min criterion for selecting an optimal transmission policy, and compare it with another natural criterion, called the total quality criterion. We present several algorithms that generate optimal transmission policies under various assumptions, including a finite client buffer, layer-encoded media, continuously scalable media, progressively rendered images, and unpredictable bandwidth environments.