EDF-schedulability of synchronous periodic task systems is coNP-hard

In the synchronous periodic task model, a set \tau_1,...,\tau_n of tasks is given, each releasing jobs of running time c_i with relative deadline d_i, at each integer multiple of the period p_i. It is a classical result that Earliest Deadline First (EDF) is an optimal preemptive uni-processor scheduling policy. For constrained deadlines, i.e. d_i <= p_i, the EDF-schedule is feasible if and only if for all Q >= 0: \sum_{i=1}^n (floor(Q-d_i)/p_i) + 1) * c_i <= Q. Though an enormous amount of literature deals with this topic, the complexity status of this test has remained unknown. We prove that testing EDF-schedulability of such a task system is (weakly) coNP-hard. This solves Problem 2 from the survey "Open Problems in Real-time Scheduling" by Baruah & Pruhs. The hardness result is achieved by applying recent results on inapproximability of Diophantine approximation.

Published in:
ACM-SIAM Symposium on Discrete Algorithms (SODA10)
Presented at:
ACM-SIAM Symposium on Discrete Algorithms (SODA10), Austin, Texas, January 17-19, 2010

 Record created 2009-10-08, last modified 2018-03-17

Download fulltextPDF
External link:
Download fulltextURL
Rate this document:

Rate this document:
(Not yet reviewed)