Protocols for the efficient dissemination of context-aware messages

Abstract

Context-aware applications are able to react and adapt to the context of their users. This context includes, for instance, location, properties of the user or their surroundings, nearby devices, etc. Over the last years, powerful mobile devices, i.e., smartphones or tablet computers, have become an important part in many people's computing life. Most of these devices maintain a continuous high-speed network connection, allowing to provide distributed applications with an uninterrupted stream of data. Additionally, a huge number of sensors, both in these mobile devices and deployed in our surroundings, enable the creation of comprehensive context models. Such large-scale context models open up new possibilities for the development of context-aware applications by providing access to relevant context information from providers all over the world. However, until now, applications need to query the context model for relevant information or register for events or messages; it is not possible to "push" information to the mobile devices, neither from the infrastructure nor from other mobile devices. To support application developers, we propose Contextcast, a novel communication paradigm that allows for the dissemination of context-aware (or contextual) messages in a system of context-aware routers. This includes the fundamental semantics to address clients using context constraints and a reference dissemination scheme for such messages. To enable Contextcast to grow to scales similar to the context-aware systems that it is intended to be used with, we also propose a couple of optimized routing approaches. They are designed to reduce the number of maintenance messages that are necessary for the dissemination of contextual messages. One optimized routing algorithm uses coarse context information to reduce the amount of context updates propagated to routers. To this end, routers use the similarity of contexts to automatically find groups of similar clients, whose information can then be propagated as a single, coarse context. While this reduces the amount of context information to be propagated, the resulting information loss causes more messages to be forwarded, since routers no longer possess exact information to match against the constraints in contextual messages. A configurable similarity threshold allows for various trade-offs between the coarseness of the context information and the resulting additional message load. The second orthogonal routing approach relies on statistics to determine the characteristics of contexts and messages in the system. Without context knowledge, routers must assume the presence of a matching recipient and forward a message speculatively to disseminate it to all recipients. Using statistics, routers can determine how often certain messages occur and then calculate the benefit of propagating contexts corresponding to these messages. Several parameters enable an administrator to adjust how fast the system reacts to changes, depending on the observed messages and context updates. Additionally, temporal support extends Contextcast with a powerful mechanism that allows application developers and clients to address messages to certain contexts in the past or future. This includes an additional context attribute \cattr{time} and a constraint with various, easy to use temporal operators. We also propose efficient routing approaches for historical and future messages. Routing historical messages focuses on efficient routing while effectively protecting the clients' privacy, i.e., their respective context history. The routing approach for future messages delays forwarding messages until a matching context is registered, thus preventing needlessly forwarded messages.