Abstract

The IP Multimedia Subsystem has seen increasing deployment over the past few years. This also means that the number of subscribers has significantly increased. Thus IMS performance analysis becomes a critical area to be researched. There are various methods to conduct the system evaluation. Most of the previous work has concentrated solely on methodology models or physical system measurements. This thesis proposes a new model combining a queueing network model and physical system measurements to achieve a precise and realistic evaluation result.

The proposed model uses the concept of an open multi-class queueing network with heterogeneous requests. The requests enter IMS system at a Poisson distribution rate and are grouped into different classes. They travel within the queueing network with or without changing classes depending on which nodes are being processed. After passing through the required nodes, the requests exit the network. The serving time at each node is measured and the final system response time can then be derived using our formula. Our model can also predict the IMS saturation point over which the system becomes unstable. In addition, the CPU utilizations for each IMS component are derived. Based on these values, multiple IMS components can be efficiently grouped onto one machine to save system resource.

The proposed model is verified for the IMS registration, call setup and termination procedures in an IMS test-bed system. The measured and calculated performance results match precisely. In addition, the model can group multiple IMS components or can separate one IMS component into multiple logical components. This scalability is crucial in production systems where the number of components grows continuously. Service providers can use this model to study the message flows in their IMS network and see how the system responds when the traffic load changes.