Improvement in Feasibility Testing for Real-Time Tasks

Abstract

We investigate the problem of preemptively scheduling Real-Time systems on a single processor. We first present two novel characterizations of feasible task sets (in our task model, deadline is allowed to be different to period). These characterizations lead to a feasibility testing algorithm. The proposed testing algorithm outperforms all previous feasibility tests. Since the feasibility problem is known to be co-NP-complete, our algorithm has a worst-case exponential time complexity. However, experimental results indicate that it runs on pseudo-polynomial time for a very large percentage of task sets (virtually, all real systems can be solved efficiently). We also show that our algorithm represents a sufficient condition for asynchronous task sets to be feasible. Our main contributions, besides the new feasibility test, are the theoretical tools that are the characterizations of feasible task sets. We hope they will enable lead to more advances in the field of scheduling theory.