With the advent of LTE networks, greater range of services becomes available. In parallel, convergence of networks and services also becomes one of the main focuses in telecommunication industry. The result is the IMS, which provides network convergence through access independence. Even greater focus is put on quality of user experience, with users expecting the services to be constantly available, so thorough testing needs to be done before the system is opened for public use. One factor for IMS performance is the ability to handle large quantities of signaling traffic, and that was the focus of this work. An application that emulates real UAs was developed and used to start sessions on an existing IMS system. Different traffic models were used, namely Poisson process, exponential and linear model, and success rates were measured. The Poisson process and exponential distribution tests show that as the number of messages per interval increases, responses become delayed, which is the main source of unsuccessful session initiations. With sufficiently long wait time, the number of truly unsuccessful requests becomes minimal. Linear distribution is used for stress testing, since the intervals are sequential, with no pause between them. This causes the system to start to throttle, since new requests are stacked upon previously unprocessed ones. Even after the number of requests per interval becomes constant, the success percentage still continues to drop.