Tham khảo tài liệu 'petri net part 7', kỹ thuật - công nghệ, cơ khí - chế tạo máy phục vụ nhu cầu học tập, nghiên cứu và làm việc hiệu quả | On the Use of Queueing Petri Nets for Modeling and 171 Performance Analysis of Distributed Systems multiple runs of the simulation and the variation of measured performance metrics from multiple tests were negligible. For all metrics the standard deviation of estimates was less than 2 of the respective mean value. The metrics considered were transaction throughput Xi transaction response time Ri and server utilization ULB for the load balancer UAS for the application servers and UDB for the database server . The maximum modeling error for throughput was for utilization and for response time . Varying the transaction mix and workload intensity led to predictions of similar accuracy. However even though the model was deemed valid at this point of the study as we will see later the model can lose its validity when it is modified in order to reflect changes in the system. Step 6 Use model to predict system performance In Section some concrete goals were set for the performance study. The system model is now used to predict the performance of the system for the deployment configurations and workload scenarios of interest. In order to validate our approach for each scenario considered we will compare the model predictions against measurements on the real system. Note that this validation is not part of the methodology itself and normally it does not have to be done. Indeed if we would have to validate the model results for every scenario considered there would be no point in using the model in the first place. The reason we validate the model results here is to demonstrate the effectiveness of our modeling approach and showcase the predictive power of the QPN models it is based on. As in the validation experiments for all scenarios considered in this section the model is analyzed by means of simulation using SimQPN and the method of non-overlapping batch means is used for steady state analysis. Both the variation of point estimates from multiple runs
