Reliable Performance Data Collection for Streaming Media Services

The recent proliferation of streaming media systems in both wired and wireless networks challenges the network operators to provide cost-effective streaming solutions that maximize the usage of their infrastructure while maintaining adequate service quality. Some of these goals conflict and motivate the development of precise and accurate models that predict the system states under extremely diverse workloads on-the-fly. However, many earlier studies have derived models and subsequent simulations that are well-suited only for a controlled environment, and hence explain a limited sets of behavioral singularities observed from software component profiles. In this study, we describe a systematic performance evaluation methodology for streaming media systems that starts with the reliable collection of performance data, presents a mechanism to calibrate the data for later use during the modeling phase, and finally examines the prediction power and the limitations of the calibrated data itself. We validate our method with two widely used streaming media systems and the results indicate an excellent match of the modelled data with the actual system measurements.