3.2. Dynamic adaptation
QoS management policy will be subject to user preferences and application-specific behaviour. Applications may find it useful to have a specification of the QoS management policy before the application starts operating. This would certainly be of value to the network for controlled allocation of resources, and makes sense in the context of trying to assure end-to-end to QoS. However, in our consideration of dynamic adaptability, the use of the application typically requires interaction with the user in order to determine its adaptation requirements, and these may not be known until after the application is running. In Ref. [12], the QoS management policy: captures the degree of QoS adaptation (continuous or discrete) that the flow can tolerate and the scaling actions to be taken in the event of violations to the contracted QoS [20].
We chose to make a separation between what “the flow can tolerate” and the “scaling actions to be taken”. We argue that the former is a property of the media and the latter is an application-specific requirement that includes interaction with the user. Flow performance specifications can be used to indicate the flow-states that are possible for a flow and can be determined by the application designer. The action to be taken on fluctuations (“violations”) of QoS is a dynamic adaptation decision and cannot be determined by the application designer a priori. It is the difference between the application designer saying, “I know what is sensible for the application”, and the user saying, “I know what is sensible for the application to do for me”. Ultimately the application's functional constraints have the final say on which flow-state(s) is (are) functionally possible, but this should not dictate how the user would like the application to behave, i.e. how adaptation should take place. Ref. [21]points out that the user requirements and the network QoS may change throughout the session and proposes that the user should be given the opportunity to make informed decisions about application adaptation. As an example, consider a remote teaching scenario and the requirements of the audio and video flows. When operating in lecture mode (main part of the teaching session), the conferencing application may tolerate relatively high delay and throughput is asymmetric, but during a question and answer session (at the end of the teaching session), low delay and jitter are required with symmetric throughput for flows.
The application must be able to assess the user preferences and available network QoS in order to make automatic and dynamic adaptation decisions.
4. The QoSSpace
Our concern is that we try and assess the overall relative compatibility between the flow QoSStates and the network QoS rather than try to evaluate the equality (or otherwise) of the absolute values of the QoS parameters for a flow-state and the measurements taken from the network. We would like our model to indicate how well the network QoS matches the requirements for any of an application's flow-states.
4.1. QoSSpace, QoSParams and QoSStates
The QoSSpace is a multi-dimensional space in which flows conceptually exist, and into which the network QoS is mapped. Flows are represented by QoSStates, and each flow may have a set of QoSStates. The dimensions of the QoSSpace are represented by a set of QoSParams. This is depicted in Fig. 2(a), which shows only three QoSParams, but any number of QoSParams are possible.