CPU reservations and time constraints: implementation experience on windows NT

  • Authors:
  • Michael B. Jones;John Regehr

  • Affiliations:
  • Microsoft Research, Microsoft Corporation, Redmond, WA;Department of Computer Science, University of Virginia, Charlottesville, VA

  • Venue:
  • WINSYM'99 Proceedings of the 3rd conference on USENIX Windows NT Symposium - Volume 3
  • Year:
  • 1999

Quantified Score

Hi-index 0.00

Visualization

Abstract

This paper presents an implementation of scheduling abstractions originally developed for the Rialto real-time operating system within a research version of Windows NT called Rialto/NT. These abstractions, CPU Reservations and Time Constraints, as described in the 1997 SOSP paper [Jones et al. 97], are intended to allow: (1) activities to obtain minimum guaranteed execution rates with application-specified reservation granularities via CPU Reservations, and (2) applications to schedule tasks by deadlines via Time Constraints, with on-time completion guaranteed for tasks with accepted constraints. The Rialto/NT scheduler differs from the original Rialto scheduler in several key respects. First, it has been extended to schedule multiprocessors--this is the primary new intellectual contribution of this work. It has been adapted to operate with operating system clock services that only provide timing interrupts at regular periodic intervals measured in milliseconds, rather than being able to schedule clock interrupts at arbitrary sub-millisecond points of time. It coexists with the existing Windows NT scheduler, allowing it to schedule time not scheduled by itself. Finally, it has been implemented in a particularly non-intrusive manner, using rather than replacing the existing Windows NT priority-based scheduler. Results presented will demonstrate that CPU Reservations and Time Constraints can be effectively implemented on multiprocessors. We will also describe the implementation techniques chosen and tradeoffs made as a result of implementing within Windows NT. Finally, we will present performance results and execution traces.