Abstract. Context-aware systems represent extremely complex and heterogeneous distributed systems, composed of sensors, actuators, application components, and a variety of context processing components that manage the flow of context information between the sensors/actuators and applications. The need for middleware to seamlessly bind these components together is well recognised. Numerous attempts to build middleware or infrastructure for context-aware systems have been made, but these have provided only partial solutions; for instance, most have not adequately addressed issues such as mobility, fault tolerance or privacy. One of the goals of this paper is to provide an analysis of the requirements of a middleware for context-aware systems, drawing from both traditional distributed system goals and our experiences with developing context-aware applications. The paper also provides a critical review of several middleware solutions, followed by a comprehensive discussion of our own PACE middleware. Finally, it provides a comparison of our solution with the previous work, highlighting both the advantages of our middleware and important topics for future research.

Abstract

Privacy is widely recognised as a significant obstacle inhibiting the adoption of context-aware applications. In order to remove this obstacle, advances are required in many areas of context-awareness research. In this paper, we address the incorporation of privacy support into context models. In particular, we present extensions to our context modelling approach that address the challenges of assigning ownership to context information and enabling users to ex-press privacy preferences for their own information. MotivationContext-awareness is receiving increasing interest as a software design approach that is appropriate for pervasive computing. Context-aware software relies on various types of context information in order to make decisions about how to dynamically adapt to meet user requirements. This information is usually derived from a range of sources, including user profiles, applications, and sensors. Some types of context information are inherently sensitive and must be protected in order to satisfy users' privacy requirements.Unfortunately, providing adequate protection for context information is extremely challenging. Context-aware systems typically contain collections of heterogeneous information with variable privacy requirements resulting from (i) differences in the sensitivity of the information, (ii) differences in users' individual privacy preferences, and (iii) changes in users' privacy preferences over time and in response to context changes. The problem of controlling access to context information is further complicated by the fact that pervasive computing environments permit looser and more dynamic couplings between people and resources, * The work reported in this paper has been funded in part by the Cooperative Research Centre for Enterprise Distributed Systems Technology (DSTC) through the Australian Federal Government's CRC Programme (Department of Education, Science, and Training). thereby invalidating the usual approaches to ownership and control of resources. In traditional computing systems, resources such as files are often assigned ownership according to who creates them, and owners control access to their resources by setting permissions. In contrast, in pervasive systems, there is often no direct link between an information source (e.g., a camera in a meeting room) and the entities that should be entitled to control the corresponding context information for privacy purposes (e.g., the people captured on camera). In addition, the relationship between these entities and the context attributes in which they have an interest in terms of privacy -which we refer to in this paper simply as an ownership relation -is often context-dependent.As a result of these problems, most current contextaware applications provide very little support for privacy. Early efforts to address privacy concerns have investigated the design of privacy-preserving location sensing systems [10] and the integration of access control mechanisms into pervasive computing infrastructure [4,2]. These solutions ...

Abstract. Context-aware applications rely on implicit forms of input, such as sensor-derived data, in order to reduce the need for explicit input from users. They are especially relevant for mobile and pervasive computing environments, in which user attention is at a premium. To support the development of context-aware applications, techniques for modelling context information are required. These must address a unique combination of requirements, including the ability to model information supplied by both sensors and people, to represent imperfect information, and to capture context histories. As the field of context-aware computing is relatively new, mature solutions for context modelling do not exist, and researchers rely on information modelling solutions developed for other purposes. In our research, we have been using a variant of Object-Role Modeling (ORM) to model context. In this paper, we reflect on our experiences and outline some research challenges in this area.