On best-effort utility accrual real-time scheduling on multiprocessors

  • Authors:
  • Piyush Garyali;Matthew Dellinger;Binoy Ravindran

  • Affiliations:
  • ECE Dept., Virgina Tech, Blacksburg, VA;ECE Dept., Virgina Tech, Blacksburg, VA;ECE Dept., Virgina Tech, Blacksburg, VA

  • Venue:
  • OPODIS'10 Proceedings of the 14th international conference on Principles of distributed systems
  • Year:
  • 2010

Quantified Score

Hi-index 0.00

Visualization

Abstract

We consider the problem of scheduling dependent real-time tasks for overloads on a multiprocessor system, yielding best-effort timing assurance. The application/scheduling model includes tasks with time/utility function time constraints, mutual exclusion constraints, and arbitrary arrival, execution-time and resource access behaviors, with timeliness optimization objective of maximizing total accrued utility while ensuring mutual exclusion constraints and deadlock-freedom. Since this problem is NP-hard, we develop a class of polynomial-time heuristic algorithms, called the Global Utility Accrual (GUA), and present two algorithm instances, namely, Non-Greedy Global Utility Accrual (NGGUA) and Greedy Global Utility Accrual (G-GUA). We establish several properties of the algorithms including conditions under which optimal total utility is accrued, mutual exclusion constraints are satisfied, and deadlock-freedom is achieved. We develop a Linux-based real-time kernel called ChronOS, extended from PREEMPT RT real-time patch. ChronOS provides a framework for implementation of a variety of multiprocessor schedulers. Our experimental studies with ChronOS reveal the effectiveness of GUA algorithms under a broad range of workloads.