Details

Autor(en) / Beteiligte

• Sivaharan, Thirunavukkarasu

Titel

Towards a generic and adaptable publish/subscribe middleware to facilitate pervasive computing applications

Ort / Verlag

ProQuest Dissertations & Theses

Erscheinungsjahr

2008

Quelle

ProQuest Dissertations & Theses A&I

Beschreibungen/Notizen

• The publish/subscribe paradigm is well known for its loosely-coupled and asynchronous communication model. The paradigm, it is argued, is also well suited for pervasive computing applications given the above characteristics. There has been a great deal of research work, particularly over the last decade, on the publish/subscribe paradigm producing many novel publish/subscribe schemes and other results in specific deployment scenarios. However the major shortcoming of this body of work is that present day platforms have not yet enjoyed wide-spread deployment in real-world applications, even though large numbers of applications are naturally event-driven. As this thesis will argue one root technical cause for the above mentioned predicament is that state-of-the art publish/subscribe middleware technologies are to a large extent intrinsically tied to a particular network type, have a one-size 'fixed' monolithic infrastructure and generally offer a fixed interaction type. Hence, they are not well placed and fundamentally overstretched to cope with ever growing heterogeneity and application diversity. To address this limitation the thesis proposes an approach that is centred on a generic and adaptable framework, separation of concerns and component frameworks. The resultant middleware (named GREEN) can: 1) gracefully deliver a wide range of custom infrastructure(s) that scale-up and down in terms of functionality, complexity and size, 2) facilitate the trade-off of a large number of competing design parameters, and 3) gracefully overcome heterogeneity. These findings are validated through a range of experiments covering heterogeneous network types (e.g. MANET, WSN, NES, WAN), device types (e.g. tiny motes, mobile devices, embedded devices and PCs) and application profiles. Significantly results show the framework is indeed feasible, does not result in one-size bloated and complex infrastructure, and the overheads introduced by the adaptive operations are found to be acceptably low in general.