Limits for the real-time simulation of video services over commodity hardware

The results from event-driven simulations with video services can present limitations because of the cost of developing accurate models. Creating a video server model precise in how streams are transmitted is generally as expensive as the construction of the real server, thus several simplifications are generally introduced to reduce the effort. In this context, the lack of accuracy could be devastating if metrics such as jitter are imprecise. One potential alternative to solve these issues is the combination of real services with Real-Time (R-T) capable network simulators. This paper evaluates R-T simulation capabilities of two of the most popular tools, NS-3 and OPNET to support the performance evaluation of video services using a commodity hardware testbed. Taking a video on demand service as a test case, we have studied the use of NS-3 and OPNET in situations where their more profitable characteristics, such as instance creation simplicity or the use of analytic traffic, could produce the best results. The performed tests show how NS-3 is able to replicate an access network topology with up to 10 MPEG-4 video streams or OPNET has been able to replicate a ring core network with up to 60 of the same streams without affecting important metrics as important as jitter. These results point out the validity of R-T simulation in the context of services that have streams as a main component if their strengths are taken into account.