Scheduling and Dynamic Pricing with Service Differentiation

摘要

In this paper we consider an Internet router providing service differentiation like in a DiffServ domain, controlled by a dynamic pricing mechanism. The system is modelled by a multi-class queue and we consider that users choose their class of service by observing the current state of the system. Each user values his service using a utility function and their decision depends on their service valuation. The way of users take their decision depending on the utility function is called a policy. One way to provide service differentiation in a ingress DiffServ router is to employ different scheduling algorithms. We propose in this paper to compare, from a pricing perspective, the two scheduler Strict Priority (SP) and Weighted Fair Queueing (WFQ) used to provide service differentiation in DiffServ. We study an optimal dynamic pricing algorithm for the SP scheduler using a Markov decision framework based on reinforcement learning. Finally we compare the provider revenue with the one obtained in [1] with a WFQ algorithm.