Due to the variation of the tasks attributes, the behavior of soft real-time systems, such as multimedia application, is becoming increasingly unpredictable. Under this circumstance, the scheduling algorithms, which depend on the tasks static attributes, cant give a usable and efficient resource allocation to those soft real-time systems. This paper presents an elastic scheduling algorithm for flexible workload. Based on sampling the total number and lost number of the task instances, this algorithm adjusts the number of task instances executing in the next sampling period to guarantee the tasks basic QoS (quality of service) and to improve the system resource utilization and concurrency in the next sampling period. This paper analyzes the model and evaluates its performance. Simulation results show that, besides improving resource utilization, this algorithm has good stability and convergence.