Colloquium Speaker

Speaker:Scott Brandt
University of Colorado at Boulder
Topic:Soft Real-Time Processing with Dynamic QoS Level Resource Management
Date:Tuesday, March 2, 1999
Time:9:30 AM
Place:Gould-Simpson, Room 701

Refreshments will be served in the 7th-floor lobby of Gould-Simpson at 9:15 AM


The performance of current desktop hardware has allowed the implementation of many computationally intensive real-time applications that have traditionally required high-performance special-purpose computers. General-purpose operating systems lack mechanisms for guaranteeing the strict timing requirements of many of these applications and real-time operating systems are too restrictive for use in general-purpose computing environments. In support of such soft real-time applications researchers are developing a variety of new programming models, middleware components, and operating system mechanisms that try to balance user, application, and system needs so as to provide the best overall performance.

At the University of Colorado we have developed a soft real-time model based on the notion of QoS Levels wherein each application is written with multiple levels at which it can operate - each level providing a certain amount of benefit to the user and having a corresponding resource cost. Based on this model we have implemented a working middleware soft real-time system on top of the Unix operating system. A centralized resource manager called the Dynamic QoS Resource Manager (DQM) uses resource information from the operating system and level information from the community of applications to balance the system load, user benefit, and available resources across the collection of applications. This system provides a natural model for soft real-time application development and separates soft real-time policy and mechanism, placing soft real-time policy under the control of the application developer and the user.

This talk will present a discussion of our system including our soft real-time application model and DQM system architecture. It will include experimental results showing the feasibility and utility of this approach, interesting results with respect to soft real-time processing on general purpose systems, and a discussion of some new soft real-time metrics for analyzing the performance of such systems.