QoS parameters may include local (end-system) resources, not just network QoS measurements. For example, on mobile systems, measurements of battery power or host load 10 and 11may also be used for making adaptation decisions.
So, our network model should be able to process information about QoS parameters, but:
•
Should not put any constraints on where that information (measurements) should come from.
•
Should be able to accept a wide range of QoS parameters, whether end-system specific or network specific.
•
Be able to cope with any timescale over which that information is measured.
2.3. Making adaptation decisions
In Section 2.1, we saw that the user preferences introduce heterogeneity and that the information required to make an adaptation decision may reflect a per-application instance view of QoS. A mechanism is required that can give an indication of the ability of the network to support any of a number of flow-states that an application instance might take. In a distributed application, there may be other application-level signalling involved before adaptation can take place. Also, the application modes may be functions of other application-specific information, so our model can not make a decision for the application, but offer QoS summaries – QoSReports – that represent a view of the relative compatibility of the network QoS and the application's flow capability. However, in general we are not aware of the following application-specific details:
•
The nature in which information from a current QoSReport must be evaluated with information from previous QoSReports in order to make flow-state change decisions.
•
The nature in which QoSReports must be evaluated with other (application-specific) information in order to make flow-state change decisions.
There may be a statistical or temporal sense in which our QoSReports have meaning to the application with respect to its current mode or flow-states. The application may have synchronisation constraints between its application flows. Such matters can only be assessed by the application. So, we chose to separate the network-related flow-state information from other application-specific information. We chose that the flow-states be expressed as QoSStates, flow-state information that is specific only to the QoS requirements and constraints of the individual application traffic flows. For example, an audio application may have several different audio encoding techniques and each would be a separate QoSState for that audio flow. Furthermore, as we do not know how the application will use the QoSReports from our model, we should simply issue QoSReports that are based on an assessment of the instantaneous QoS of the network – a snapshot in time. This is also consistent with our requirement for our model to work within the timescale constraints of the application.