A context-aware reflective middleware framework for mobile ad-hoc and wireless sensor networks.

摘要

In smart environments, numerous devices need to be dynamically connected to form a Distributed Real-time and Embedded (DRE) system based on Mobile Ad-hoc NETworks (MANETs) or Wireless Sensor Networks (WSNs) and collaboratively react to changing contexts with dependable quality of service (QoS). Traditional middleware platforms, which have been designed as monolithic static systems, cannot effectively support the flexible and dynamic computing environments for emerging DRE applications. In consequence, there is an urgent need to provide a powerful adaptation approach for existing middleware.;Context-Aware Reflective Middleware (CARM), which supports dynamic reconfiguration and distributed behavior synchronization of component-based applications, has been an appealing technique for DRE systems in MANETs and WSNs. Existing CARM frameworks use single component-chain based architecture and synchronous synchronization protocols that are inefficient since they impose dependence restrictions and reconfiguration overhead. The achieved reconfiguration time is in a range of several seconds or even tens of seconds. We argue that they can not satisfy the efficiency requirements of some DRE applications in the dynamic environments, where reconfiguration is triggered every second or millisecond. Furthermore, there is no CARM framework implemented for extremely resource-limited wireless sensor nodes due to the complexity and overhead.;The key contribution of this dissertation research is the design and realization of a context-aware reflective middleware framework, called MassWare (Mobile Ad-hoc and Sensor Systems' Middleware), to meet the efficiency requirement of such adaptive DRE applications in MANETs and WSNs. Our thesis is that the reconfiguration efficiency can be improved by asynchronous synchronization support via a middleware framework. To prove this thesis, we propose a multiple component-chain based middleware architecture and an active-message oriented asynchronous synchronization protocol for the reconfiguration. The key idea behind the protocol is that each application-layer data packet takes an active message header that indexes the correct component-chain of the packet receiver to process the data payload. Therefore, the distributed behavior synchronization time is dramatically reduced by eliminating the operation suspension time and buffer clearance time. Based on the protocol, we have developed MassWare in MANETs and WSNs that helps the DRE applications adapt to changing contexts in an efficient and robust way according to user-defined adaptation rules.;In this dissertation, we describe the complete architecture design, model analysis, and implementation of MassWare, which addresses the major challenges of existing CARM frameworks: improving reconfiguration efficiency, realizing CARM in WSNs, and offering a unified development model for both MANETs and WSNs. MassWare and supported applications have been implemented on PDA platforms and Mica sensor nodes. The reconfiguration efficiency has also been analyzed and compared with those of peer CARM frameworks based on a novel theoretical model. Quantitative empirical results show that the reconfiguration time of MassWare for MANETs is reduced from seconds to hundreds of microseconds. Evaluations demonstrate that MassWare is robust, scalable and generates a small memory footprint.