A simulation-based approach to test the performance of large-scale real time software systems

Sammanfattning: Background: A real-time system operates with time constraints, and its correctness depends upon the time on which results are generated. Different industries use different types of real-time systems such as telecommunication, air traffic control systems, generation of power, and spacecraft system. There is a category of real-time systems that are required to handle millions of users and operations at the same time; those systems are called large scale real-time systems. In the telecommunication sector, many real-time systems are large scale, as they need to handle millions of users and resources in parallel. Performance is an essential aspect of this type of system; unpredictable behavior queue cost millions of dollars for telecom operators in a matter of seconds. The problem is that existing models for performance analysis of these types of systems are not cost-effective and require lots of knowledge to deploy. In this context, we have developed a performance simulator tool that is based on the XgBoost, Random Forest, and Decision Tree modeling. Objectives: The thesis aims to develop a cost-effective approach to support the analysis of the performance of large-scale real-time telecommunication systems. The idea is to develop and implement a solution to simulate the telecommunication system using some of the most promising identified factors that affect the performance of the system. Methods: We have performed an improvement case study in Ericsson. The identification of performance factors is found through a dataset generated in a performance testing session, the investigation conducted on the same system, and unstructured interviews with the system experts. The approach was selected through a literature review. Validation of the Performance Simulator performed through static analysis and user feedback received from the questionnaire. Results: The results show that Performance Simulator could be helpful to performance analysis of large-scale real-time telecommunication systems. Performance Simulator ability to endorse performance analysis of other real-time systems is a collection of multiple opinions. Conclusions: The developed and validated approach demonstrates potential usefulness in performance analysis and can benefit significantly from further enhancements. The specific amount of data used for training might impact the generalization of the research on other real-time systems. In the future, this study can establish with more numbers of input on real-time systems on a large scale.