10 likes | 139 Views
Open Sensor Web Architecture: Integration of Sensor Networks and Grid Computing - Rajkumar Buyya, Xingchen Chu and Jiye Lin.
E N D
Open Sensor Web Architecture: Integration of Sensor Networks and Grid Computing - Rajkumar Buyya, Xingchen Chu and Jiye Lin The Open Sensor Web Architecture (OSWA) proposed by NICTA, at University of Melbourne extends the Sensor Web Enablement (SWE) and further integrates the Sensor Web and Grid Computing as well as providing middleware support of Sensor Web. The Sensor Collection & Registry Services have been implemented and deployed on Tomcat and Apache AXIS. Clients can enquiry the Sensor Registry Service about available Collection Service according to their requirements and send data query via SOAP messages to obtain the observation data conformed to the XML schema defined by Observation & Measurement. Different type of connectors have been designed to fit into different kind of resources including sensor networks running on top of TinyOS, TinyDB application and remote observation data archives. Figure 1- Layered Architecture A temperature monitoring application written by NesC has been developed and deployed onto Mica2 Sensor Motes provided by Crossbow’s sensor development Kit. The Collection Service collects the sensor data, converts them into O&M XML format and then stores the data into the repository using JDO persistent mechanism. Also a simple GUI client has been used to send queries to the Collection Service via web services calls. Evaluation of Performance on both auto-sending and query-based mode have been conducted on real-sensors as well as the simulation environment TOSSIM that runs TinyDB application. Figure 2 Sensor Web Collection Service Architecture Figure 4 Performance of collecting auto-sending and query data Figure 3 GUI Client for Collecting Sensor Data The performance for the auto-sending mode application is quite moderate when the number of clients who request the observation simultaneity is quite small. Even the number of clients reaches 500, the response time for small number of records is also acceptable. In contrast, the performance of query-based model is fairly unacceptable even just one client requesting one observation takes 34 second. The response time increases near linearly when the number of clients and the number of records go up.