By analyzing preemptions of real-time tasks in embedded systems under preemptive EDF (Earliest Deadline First) schedule algorithm

we create a preemption model of periodic real-time tasks.This model describes relationship between overhead occurred by preemptions and the properties of real-time tasks

such as priorities

periods

execution times

deadlines

release times etc.It also shows the relationships among task's properties

preemption orders

schedule overheads and schedulibility.Underlying the model

this paper presents an improved preemptive EDF algorithm

which uses the release times as schedule criteria got by an optimization algorithm.The optimization algorithm is based on genetic algorithm and runs off-line on desktop computers.Using the algorithm can reduce the number of preemptions

change order of preemption

so that the schedule capacity and real-time performance of embedded systems are improved.And then the improved schedule algorithm underlying the model is evaluated by experiments.