5G-Xcast is proud to announce that deliverable D5.3 on Application Layer Intelligence is now released and can be downloaded from this link.
As a central principle to the approach to exploiting multicast, we stipulate that service requirements are preferentially met without the assistance of special features provided by the network operators, particularly where such features would add complexity to the technical and commercial relationships between different organisations. For example, adaptive streaming is preferred over network QoS, despite the fact that a QoS-based solution could be considered technically superior to adaptation over a best efforts channel.
Note that this principle is not something that a network operator would necessarily advocate. It is rather a pragmatic observation that it is difficult to monetise network technology if it results in more complex cross-organisational interfaces and lower flexibility to test or introduce new capabilities. When developing network solutions we need to take into account that content service provider would rather deploy technology at protocol layers that they have under their direct control.
In this document, we loosely refer to the collection of such techniques that could be applied at higher protocol layers (above layer 3) as ‘application-layer intelligence’.
The objective of this document is to identify possible application-layer techniques that might be used so that we can understand whether the point to multipoint content delivery framework will prevent these techniques from operating correctly. Conversely, we will also consider whether the application-layer techniques will prevent the framework from operating effectively.
Many of these techniques concern the dynamics of Quality of Experience control. Within the video encoders, within the adaptation algorithms that run in streaming applications, and even in the TCP behaviour itself, there are semi-independent control loops which can be argued to be optimising their own limited view of QoE. The impact of introducing multicast will be to change the dynamics of the server to client connection in such a way that is very likely to impact the effectiveness of these QoE optimisation techniques. Conversely, the use of such techniques could render the use of multicast less effective than it might otherwise have been.
Another impact to consider relates to trust and security. For the network operator to be able to insert proxies en-route and to gather enough information about a content service to steer traffic appropriately, it will need to have more access to the content stream than would normally be the case in a traditional CDN architecture.
After examination of a number of application layer techniques, we conclude that most still have value when multicast is used. Some may impact the design of the multicast system. For example, careful selection of rates will be necessary to avoid confusing adaptation algorithms.
Content protection poses no problem at all, however transport layer protection does need a specific agreement between the CDN operator and the network operator.
Stay tuned for more deliverables to come soon!