Abstract
This paper presents an ACO algorithm to search for feasible schedules of n real-time tasks on M identical processors. Unlike existing works, the proposed algorithm addresses the problem of preemptive scheduling rather than non-preemptive scheduling. A learning technique is integrated to detect and postpone possible preemptions between tasks. The proposed learning technique is also used to develop a necessary condition for the schedulability of the input task set. Experimental results show a significant success ratio improvement of the proposed scheduling algorithm.