home · mobile · calendar · colloquia · 1998-1999 · 

Colloquium - Kravets

Cooperative Solutions to the Dynamic Management of Communication Resources
Georgia Institute of Technology

Support for large-scale distributed interactive applications demands resource management infrastructures that understand the relationship between application requirements and the services available from the network. To provide such support, we have designed and implemented an end-to-end communication layer which mediates between an application and a network service. By using a dynamic communication framework, the communication layer can adaptively configure an application's communication parameters as demanded by changes in both service requirements and resource availability. Adaptation decisions maximize the perceived benefits a specific communication configuration offers to the application, by permitting applications to provide dynamic service quality requirements to the communication layer in the form of "payoff functions". As a result, the communication layer supports precisely the cost-quality tradeoffs the user is willing to make.

The effectiveness of these techniques is demonstrated with three different cost-quality tradeoff scenarios. The first scenario explores the use of a variable reliability framework that realizes reduced transfer delays by exploiting tradeoffs with respect to the reliability of message communication. The second scenario involves power management for mobile computers, including techniques for saving energy during active communication, as well as during idle periods. The communication layer exploits certain aspects of mobile communication protocols to understand the tradeoffs between energy consumption and transfer delays. The third scenario adapts communications performed within distributed virtual environments. For the multiple simultaneous data streams used in such environments, the communication layer must consider the effects of changes in the quality of one stream on the quality of other streams, as well as the effects on the performance of the application itself.

Hosted by Evi Nemeth.

Department of Computer Science
University of Colorado Boulder
Boulder, CO 80309-0430 USA
May 5, 2012 (14:13)