“QoS” is sometimes thought of as something that happens inside a “Diffserv domain”; packets are “marked” by special routers at the edge of the domain and then prioritised accordingly by the high-speed routers in the network’s core. Network managers may think that they’ve done a good job of implementing QoS when such a system has been set up and configured according to the prevailing policy.
Unfortunately such an approach is likely to miss the most significant source of QoS problems for the end user – the access network, otherwise known as “the last mile”. For the vast majority of users, the speed of their connection to the network is far lower than the capacity of their end equipment to generate and consume network traffic. As a result, the last mile link is a major – perhaps the largest – point of contention for most users’ network services. Managing the effects of contention is what QoS is all about, so applying QoS in the core while omitting to do so on the access link is to miss the point.
Some operators think that there’s no need for QoS on the access link because there’s only one subscriber there. Nowadays, though, most household broadband connections are shared by the whole family, and business broadband is obviously used by the whole office. Even where there is only one user, however – as might be the case with a mobile broadband connection – they can be expected to be using more than one networked application; making a VoIP call, for example, while viewing a web page and up/downloading email in the background. Failing to differentiate the QoS for these different applications over the access link will lead to trouble, churn, and demands for ever higher bandwidth in the hope this will alleviate the problem.
Managing access link QoS is not about extending the Diffserv domain to the end user; Diffserv markings from end equipment are unreliable, so a different approach is needed. In any case the Diffserv standard only supports one class of real-time traffic, which is inadequate for the variety of rich multi-media applications available today. A more dynamic and fine-grained QoS mechanism is needed, specifically to optimise the use of the slowest part of the network for multiple real-time, near-real-time and other critical applications; once that has been done, core network QoS can take over.