Speaker: Daniel Mosse
Affiliation: University of Pittsburgh
Date: 11/16/2000
Time and Location: 4:15 PM, B17 Upson Hall
Title: Towards a View of Efficient Softer Real-Time

Much of the research in real-time systems revolves around so-called hard real-time systems, in which missing a deadline causes catastrophic consequences.  These systems typically assume stringent requirements and very well defined task characterization; the violation of these assumptions cannot lead to deterministic guarantees. Due to the hard nature of the problem little work has been published in the realm of guaranteed computations in soft real-time systems.

This talk will present some results of real-time systems that revolve around guaranteeing softer deadlines while enabling the system to reap the benefits of dynamic analysis of the extra time that becomes available when tasks in the system do not perform as expected.  In particular, some new algorithms in imprecise computations, multiprocessor scheduling, reward-based computing, and power-aware real-time scheduling will be presented, as well as quantitative results from our simulation studies and some results from actual implementations on Linux and MACH operating systems.  The conclusions from this work leads the way to our research in adaptive, softer real-time research.