Integrated Services Model

The Internet is not a static set of services and protocols, and there has been a great deal of effort since 1990 to add a broader range of services to the Internet model. The Integrated Services Working Group [Braden, 94] of the IETF has now defined 5 classes of service which should math the vast majority of future applications (although the scheme is extensible, so that future applications which need new services are not excluded!).

The current five classes of service are:

1.

Best Effort - this is the traditional service model of the Internet, as described above, typically implemented through FIFO queuing in routers.

2.

Fair - this is an enhancement of the traditional model, where there are no extra requests from the users, but the routers attempt to partition up network resources in some fair share sense. This is typically implemented using a random drop approach to overload, possibly combined with some simple round robin serving of different sources.

3.

Controlled load - this is an attempt to provide a guarantee that a network appears to the user as if there is little other traffic - it makes no other guarantees - it is really a way of limiting the traffic admitted to the network so that the performance perceived is as if the network were over-engineered for those that are admitted.

4.

Predictive or controlled delay - this is where the delay distribution that a particular flow perceives is controlled - this requires the source (or a group where it is applied collectively to all sources sending to a group) to make some pre-statement to the routers that a particular throughput is required. This may be rejected.

5.

Guaranteed - this is where the delay perceived by a particular source or to a group is bounded within some absolute limit. This may entail both an admission test as with 3, and a more expensive forwarding queuing system.

The separation of these service classes is important, since the billing model of the network is related to the service model. For example, elastic services such as those we have traditionally used in the Internet do not require a usage charge for traffic which gets no guarantees. However, when an application needs, or asks for guarantees, there is a requirement to present some feedback to prevent everyone idly asking for the maximum guarantee (so that the network can make an informed decision). This feedback can most easily be provided by billing, although some researchers assert that it is only necessary to actually incur a charge when the network would be unable to meet all the current requests, rather than whenever people make a request. This is analogous with billing people for road use during congested periods, and not at other times, and billing people with larger cards more so as to adjust the demand.

This aspect of the Internet is relevant to considerations of videoconferencing, since it may well be that large parts of the Internet will not permit such applications until either reservation, or billing or both are in place as new technology.