Weighted Fair Queuing (WFQ)

n:PROPERTIES: :ID: 230C477D-E7C3-4F30-9A00-82FA80406696

:END:

A Queuing model used for Packet Scheduling, where, like Round Robin Queuing, Packets arriving in the queue are classified into priority-classes and the scheduler will server the classes in a circular manner. It will thus transmit a packet from class A, then class B, etc. and repeat the pattern when it has transmitted from every class.

It is a work-conserving queuing discipline and will skip classes with an empty queue.

Each class gets a different fraction of the transmission rate. Each class $i$ is assigned a weight $w_i$. During any interval of time, during which class $i$ have packets queuing, that class is guaranteed to receive a fraction of service equal to $w_i/{\sum }_{j\in \mathcal{S}}{w}_j$ where $\mathcal{S}$ is the set of classes that have packets queued.

So for a link with transmission rate $R$ class $i$ will have a throughput of at least $R\cdot w_i/{\sum }_{j\in \mathcal{S}}{w}_j$.

This analysis is however idealised and does not account for the fact that packets are discrete and that a transmission will not be interrupted for a transmission of another packet.