Graduate Students: Seok-Kyu Kweon, Emmanuel Abram-Profeta, and Sung H. Choi
Faculty: Kang G. Shin
Sponsor: NSF
Most multimedia applications require a predictable communication subsystem. At the current stage of our research, we focus on three problems. The first direction is to develop schemes for multiaccess networks which can provide performance guarantees. We have developed a real-time communication scheme for multiaccess links by using a link control unit to establish real-time channels. In a multiaccess link environment, a real-time channel of certain grade of quality may be implemented by allocating sufficient link bandwidth to the source node of the channel. The system also takes advantages of multiplexing independent statistical channels. Without compromising performance requirements, we have successfully reduce the link capacity that needs to be reserved for a channel to the average level from the worst case level.
The second problem is to develop a distribute route-selection strategy for point-to-point networks. We want to find a route which meets the end-to-end delay requirement at that moment while minimizing the time for channel establishment. We also do not want to decline a request if a such route exists. By measuring the minimum feasible worst-case delay of each hop, and applying the Bellman-Ford algorithm, we may obtain a route of the requirement if it does exist. If only a few real-time applications are considered, this route selection strategy can be significantly improved by maintaining a table. With these tables, the route selection problem can be solved by a table lookup.
Real-time multimedia applications require guaranteed performance communication services, such as throughput, delay, delay jitter and loss rate. In guaranteed service models such as the real-time channel in point-to-point networks, or the statistical real-time channel in multiaccess networks, resources are reserved within the network. The characteristics of the source, such as peak and average rate, are supposed to be known at channel establishment time. This assumption does not hold with live VBR video sources. In this situation, the channel parameters have to be predicted, and, if possible, updated at run-time. Our approach is to develop an observation-based algorithm to forecast statistical characteristics of live video sources, and to predict the channel parameters for different specification models (such as the real-time channel model). We will validate the study with simulations and tests of goodness-of-fit of predicted models with real distributions taken from MPEG, motion JPEG and H.261 video data.
Many multimedia applications do not require lossless transmission. category. As B-ISDN is being developed, future multimedia applications will use wide-area, packet-switched networks. In this context, one direction of our work is to develop algorithms to provide statistical guarantees for loss-tolerant channels, in point-to-point networks. Our goal is to achieve higher network utilization and statistical multiplexing gain while guaranteeing quality of service.