This paper investigates the feature interaction problem, known from traditional telephony environments, in the context of Internet personal appliances (IPA). IPAs are dedicated consumer devices which contain at least one network processor. They include the Internet alarm clock, which takes into account road conditions or expected arrival times of air planes when setting the alarm time, or the Internet enabled fridge, which keeps an inventory of groceries and issues orders to suppliers. The results of our investigation are threefold. The first part of this paper introduces a service taxonomy supported by a list of example services. The second part discusses feature interactions between services for appliances. A classification for such interactions and example interactions and possible conflicts are presented. The final part contains an outline for an approach to handle such interactions.

This document proposes the use of session initiation protocol (SIP) for network-capable appliances. It leverages standard SIP capabilities to directly communicate with appliances even when they are behind firewalls, network address translators (NATs) or other entities that prevent direct end-to-end communication. When combined with the recently proposed instant messaging and presence SIP extensions these techniques become even more powerful to allow the secure and reliable access to appliances in a variety of different networked environments.

The Intelligent Network (IN) allows rapid changes in the services provisioned and their functionality. Services may be supplied by different service providers, making it unlikely that all service specifications will be available for examination by any single agency. Approaches to handle feature interaction problems must be able to operate within these constraints. Work by the authors has produced a generic run time feature interaction manager (FIM) concept to manage feature interactions in a live network. It monitors features as black boxes, learns their "correct" behavior and uses this to determine when feature interactions have occurred. The paper describes and compares experiences using three different techniques to realize the proposed approach. These are: state sequence monitoring, artificial neural networks (ANN), and rule based monitoring which also includes integrated generic resolution approaches. The paper explores the design alternatives with the various techniques, and reports on the results obtained from experimentation.