Server link load modeling and request scheduling for crowdsourcing-based benchmarking systems

Crowdsourcing-based network benchmarking is a novel method of large-scale performance evaluation, which allows for the collection of measurements from end nodes in order to estimate overall network performance. For accuracy and fairness, the measurement server and its communication links must not constitute the bottleneck in the overall benchmarking system. In this paper, we focus on performance optimization through efficient request scheduling. We present a modular approach to model the load on the server-side communication links analytically, allowing for the combination of simulation parameters with real measurement results, and demonstrate the potential benefit of scheduling measurement requests with a discrete time-based approach. We show that efficient scheduling algorithms are as important as hardware improvements (such as upgrading the communication links to higher capacity), in order for benchmarking systems to cope with the increasing number of measurements and ever growing data rates in both fixed and mobile technologies.