|This paper presents a novel approach to the automated detection
and resolution of Feature Interactions during runtime using techniques
borrowed from Transaction Processing theory. A set of algorithms is presented
which allow a feature to 'suggest' responses to an event from Call Processing.
A Feature Manager is then used to accept or reject these responses and
optionally to 'roll back' the feature to a previous state where is has
never seen the stimulus so that interactions can be avoided. When multiple
features respond to an event, this technique can be used to build up trees
of possible posterior states from an initial event, which can then be parsed
to determine a suitable resolution to the interaction.
The rollback technique opens significant new opportunities where a Feature Manager (or even the features themselves) can 'experiment' with their responses to stimuli to achieve a preferable result.