Approximate Fair Queueing: A Low Complexity Packet Scheduler for Embedded Networks

Fair queueing is a well-studied problem in modern computer networks. There are two classes of queueing algorithms. One focuses on bounded delay and good fairness properties. The other focuses on the performance of the scheduling algorithm itself. However in embedded networks working under real time constraints, equally important is the deadline imposed by the application. Modern queueing algorithms do not address that fact explicitly. In this paper we propose a scheduling algorithm, Approximate Fair queueing, that aims to bridge this gap. Approximate Fair Queueing schedules packets based on the packet’s deadline. In doing so, it maintains fairness in allocating resources to all competing flows. The delay that a packet experiences due to the execution of the scheduling algorithm is also reduced by reducing the frequency of execution. We perform extensive evaluation by simulating a system with a single bottleneck node and up to 1000 concurrent flows.