The rate-monotonic (RM) algorithm is a classic fixed priority real-time scheduling algorithm. However, in most embedded real-time systems where the workload is composed of many tasks of high frequency and short execution time, the overheads from preemptions of the real-time operating system will lead to a low resource utilization rate if the RM algorithm is directly used. By studying the preemption relationship of the tasks scheduled by RM algorithm, a model of preemption overheads is established with task attributes, based on which the run-time preemption overheads of RM algorithm are reduced by optimizing the start time of the tasks in embedded real-time systems. Finally, the experimental results show the validity of the proposed strategy.