Recent research efforts in the design of both communications networks and applications have led to increased adaptability in both domains. Flexible networks support a large variety of applications efficiently and facilitate the introduction of new applications. Adaptable applications can use a wide variety of networks, such as wireless, local-area, and Broadband Integrated Services Digital Networks (BISDN's), without modification. Surprisingly little research has focused on the interface between applications and networks, however. Currently proposed interface models often are poorly defined or so simple as to hinder high application performance and efficient network resource use. This thesis shows the feasibility and benefits of a richer channel setup interface by presenting a new interface model and then showing how video applications and networks could use the model to provide high-performance service with high transport resource utilization. First, we propose Medley Interface model, which combines substreams' data rates, delays, and loss rates, and further allows bounds to be placed on the burstiness or spacing of substreams' data rates, delays, and loss rates, and further allows bounds to be placed on the burstiness or spacing of substreams' losses. Loss burstiness control is beneficial to applications such as video or file- transfer whose performance varies as much with their channels' loss spacing as with their loss rates. Next, the thesis presents a channel parameter negotiation method that reduces network resource requirements while maintaining a constant level of application performance. This iterative minimization technique achieves channel cost reductions ranging from 20% up to 70% with several applications; these negotiations require the detailed transport description provided by substream decomposition and the Medley Interface QOS format. We present new variations of existing video coding algorithms that maintain good video quality over the range of channel parameters that might result from negotiations. Leaky motion compensation causes transmission errors to disappear quickly and smoothly. When performed adaptively based upon the coded scene's contents, the bit-rate penalty of leaky compensation can be made very small. Finally the thesis presents several new buffer access disciplines that allow networks to provide channels with highly correlated or widely separated losses. These special-purpose disciplines allow networks to allocate as much as 50% less buffer space as would be needed with generic buffer disciplines. Together, the network interface, video coding methods, and buffer management disciplines presented in this thesis show the benefits and feasibility of a richer call setup network interface than has been envisioned. Video application can operate with a range of channel QOS parameters, but they must have some control of the parameters to adapt to produce video with high subjective quality after transmission. Network can provide channels with delay, loss rate, loss priority, and loss spacing characteristics finely tuned to the needs of specific applications, but these needs must be made known to the network. These channel characteristics are specified via the Medley Interface's substream decomposition and new QOS format; further, these components facilitate channel setup negotiations that minimize an application's transmission cost at a fixed performance level.




Download Full History