Many telecommunication applications, such as multimedia streaming platforms and connected devices (e.g., smart vehicles and telemedicine), see their Quality of Service (QoS) affected by their allocated buffer size. Indeed, whenever the data traffic volume of a user gets larger than the allocated buffer size, a decrease of the service quality is experienced. Knowing that increasing the buffer size usually incurs extra costs, a relevant optimization problem is to find the optimal buffer size so that the amount of wasted data traffic and QoS mismatch are minimized.

To tackle this problem, the authors propose a static approach as well as a dynamic one. The static approach consists of finding the best buffer size partition among a given set of users, while a vector of expected data volumes to be processed during a fixed time horizon is given for each user. The buffer size is considered to be the same for all users and constant throughout the planned time horizon. This amounts to solving a nonlinear optimization problem which is performed with Artelys Knitro. The dynamic approach is based on a discrete-time linear time-invariant Model Predictive Control (MPC) formulation and focuses on determining the buffer size allocated to each user, which can vary along the planned time horizon allowing to enhance the user-perceived QoS.

Numerical simulations are conducted to evaluate the proposed approaches and the computational results demonstrate their effectiveness both in terms of buffer savings and QoS mismatch decrease. For instance, the static approach achieved with Artelys Knitro was capable of dividing by 6 the buffer requirements leading to important savings while providing a QoS similar to the legacy practices typically proposed by most mobile network operators.