The performance of scheduling algorithm is affected remarkably by the processing capability of processors and
the size of tasks. Therefore, a task-processor model is proposed to describe the load status of the multi-processor systems,
which reflects whether the tasks to be processed exceed the processing capability of the systems. A prerequisite describing the
schedulability of multi-processor systems is presented, and a method is designed to generate the random task set. Based on the
thrift algorithm, load-balancing thrift algorithm is proposed, to schedule multi-processor systems, which can simultaneously
increase the success rate of scheduling, reduce the average response time of tasks and schedule length, and improve the utility
of the processors evenly.