On the optimality of size-based scheduling in networking

Abstract

In recent years flow scheduling on the Internet has attracted a lot of interest in scientific research, in particular the study of how the distribution of flow size can influence system performance. Many queuing models have been studied and designed to prove that size-based schedulers improve performance for small flow without degrading overall system performance. On the other hand, however, it has been demonstrated that it is not easy to identify small size flows. In this thesis we propose a new queuing system model, starting from the study of existing ones, with a multiple level scheduling that can separate small flows from bigger ones in order to prioritise them. We show the mean response time of each queue and of the system and we compare it with the systems already present in the scientific literature. Finally we propose an idea to implement the model in the reality analysing some schedulers already present in Linux systems.