Research On Transaction Processing In Multiple Data Streams Management System For Sensor Networks

With the developments and advances in communication technologies, embedded computation technologies and sensing technologies, the integrated low-power sensing devices have attracted a considerable worldwide attention in recent years. The wireless sensor networks (WSNs) will permit remote object monitoring and tracking by deploying many sensors in a wide range of sensing area where the phenomena of interest may appear, collaboratively gather and process information of the phenomena, and report desired information to observers. Consequently, WSNs have arisen a great revolution to our life styles by integrating physical-world and information-world as a closely coupled component, and placed us at the doorstep of a new era where small wireless devices will provide access to information anytime, anywhere, and avtively participate in creating smart enviroments. WSNs are being developed for a variety of civil and military applications, such as object tracking, infrastructure monitoring, habitat sensing, and battlefield surveillance.Contrasted with other traditional networks, WINs are well characterised by some special features, such as low-power supplying, limited computation and communication resources, large scaled deployments, dynamic network topologies, mounts of rapid data streams, which bring us many new research challenges in the fields of both fundamental theories and engineering technologies.State-of-the-art techniques for WINs focus mainly on simple data-gathering applications, and in most cases, support one application per network. Therefore, the design of the network protocols and applications are usually closely-coupled, or even combined as a monolithic procedure. However, such procedures are sometimes ad hoc and impose direct interaction with the underlying embedded operating system, or even the hardware components, of sensor nodes. It is envisioned that the development of WSNs will finally demand a systematic application design based on standard and portable abstractions of the system. Thus, middleware sitting between the network hardware, operating systems, and network stacks and the applications is required to provide a runtime environment that can support coordinate concurrent applications and standardized system services to diverse applications to achieve adaptive and efficient utilization of system resources.Traditional distributed middlewares, such as DCOM, CORBA, PVM, MPI, are normally heavyweight in terms of memory and computation requirement and therefore not suitable for WSNs with scare energy and processing resources. Moreover, the WSNs middleware design needs to address the unique operating modes of WSNs that are significantly different from traditional networks, including the ad hoc deployment, untethered operation, continuous execution, and dynamic operating environments. Therefore, adaptable, easily implementable, lightweight middleware design is desired for wireless sensor networks.Several design principles for WSNs middleware are first identified. These principles motivate us to design a cluster-based multiple data streams management system (ARTs-MDS) that takes WSNs as a distributed database system integrated in the system. The system is capable of providing a virtual machine abstraction to separate application semantics from the underlying infrastructure; supports coordinate concurrent applications and effects efficient tradeoffs between the multiple QoS dimentions of an individual application as well as multiple applications.As the procedure of data stream processing is continuous that differs from the phasic execution in traditional distributed systems, the collaboration among task instances is under the stringent restraints of logic consistency and time consistency. Therefore, how to ensure logic and time consistency between continuous coordinate tasks and how to achieve an atomic global objective are the main research issues for data stream management system. In ARTs-MDS, we propose a two-layered task collaboration mode to adapt the data stream processing in sensor networks. Normally, sensing application will be triggered by a complex event, which consists of a set of primitive events, each of which will activate a sensor cluster to collaboratively gathering information about the primitive event. Tasks processing the collected data from various clusters can interact to achieve a desired global objective not achievable by any single one cluster in a coordinate style.By using the concepts of sphere of control and techniques of transaction, we address an adaptive nested transaction model to solve the key problems above. The main idea behind the model is that the system can form a dynamic sphere of control, which is considered as a transactional unit, for a set of task instances that are time consistent and logic consistent. Thus, we can meet the requirements for coordinate concurrent tasks and efficiently store the derived data by guaranteeing loosen ACID features, which are specially defined for coordinate concurrent tasks in sensor networks, of the transaction. Then, we educe the performance rules of the transaction based on the ACID definition, and prove the model adaptable to WSNs by a construction method.We implemented a task menegement sub-system in ARTs-MDS based on the model, in which, a transaction manager, described using a C-alike language, performs as a coordinator to control other compenents working collaboratively. We also design a two-layered task scheduler based on EDF and FIFO methods.Our performance evaluations show that by applying the system, the export transmission ratio of the collected data can be significantly raised and energy consumption can be reduced versus to centralized sensor fusions.