It has been stated from the early chapters that IP networks will be the target platform for this work. Therefore, this decision deals with the modelling of the network service rather than the selection of the network technology itself. Maintaining consistency with the previous chapters, QoS will be the main criterion for this decision.
As far as it concerns throughput, the current network service is considered generally adequate, provided that H.263 is used. The latter, is designed to provide video services over PSTN-level bandwidths, so the access network is less than an issue. However, from the QoS point of view, the service is definitely inadequate, without any minimum or average throughput guarantees. A solution could be the adoption of Internet multimedia architectures, however, it requires a considerable effort to implement them in all the Internet routers.
Similar remarks can be made for the delay requirements. Moreover, depending on the error control techniques that will be selected in the next paragraph, the applications will experience different latencies. For example, if interactive error control is used, increased delay is one of the main consequences, and the design should target its minimisation. Note that at this stage, decisions are taken concurrently, according to the model depicted in Figure 5.1.
The simplest approach to model error rates in the Internet is apply random packet loss. This can be arguably simulated as a random drop-out of packets just before the decoding. However, the use of techniques such as RSVP to allocate resources and provide a level of QoS makes the simulation of packet loss much more complicated, as routing decisions are taken to minimise error rates. Modelling packet loss is alone a field of ongoing research, with divergent results. Therefore, and for the sake of simplicity the initial random model is usually applied, and has been selected for this work.
Multicasting is currently implemented in `island' subnetworks but is well on the process of extending everywhere on the Internet. For open-loop encoding,5.3 multicast connections can be effectively simulated by unicast equivalents for this work's purposes. However, for interactive sender-receiver encoding and error control schemes, congestion control measures have to be taken to handle the feedback traffic effectively. The requirements of such schemes were considered well beyond the timeline and scope of this project, and were not implemented.