中文摘要：

为分析和评估多租户应用的资源消耗并预测其未来对资源的需求，研究了从细粒度上改进多租户应用性能建模的方法，在使用排队Petri网刻画应用系统能力的基础上，综合考虑工作负载、基础通信流和服务时间等因素，利用Kalman滤波快速适应参数变化的特征，建立优化的多租户应用性能模型。仿真实验表明，建模评估结果与实际性能度量值相比具有可接受的误差值，能较好地反映实际多租户应用的运行状态和预测评估各租户应用的资源需求。

英文摘要：

To analyze and evaluate resource consumption of multi-tenant application and predict the future demand for resources, the improved method of performance modeling in fine-grained way for multi-tenant applications is studied in the paper. On the basis of using the queue Petri net to de-scribe the capacity of application, taking workload, basic communication flow and service time into consideration, an optimized model is built based on Kalman filtering that can adapt to parameter var-iation rapidly. The simulation results show that the error value between the modeling results and the actual performance metrics is acceptable, so as to reflect the actual running state of multi-tenant ap-plication better and predict resources demand of each tenant application.