Graduate Exam Abstract

Tarun Banka

Ph.D. Final

May 15, 2007, 2-4pm

Engr B101

Application-Aware Transport Services for Sensor-Actuator Networks

Abstract: <br> Many emerging mission-critical sensor actuator network applications rely on the best-effort service provided by the Internet for data dissemination. This dissertation investigates the paradigm of application-aware networking to meet the QoS requirements of the mission-critical applications over best-effort networks that do not provide end-to-end QoS support. An architecture framework is proposed for application-aware data dissemination using overlay networks. Using the proposed architecture framework, an overlay network based application-aware one-to-many high-bandwidth protocol is developed that performs application-aware processing at overlay nodes in the best-effort network to meet the QoS requirements of the heterogeneous end users. Some of the examples of application-aware processing at overlay nodes include application-aware rate adaptation during congestion control, and selective packet forwarding/drops within the network. An application-aware congestion control protocol performs data selection and real-time scheduling of data for transmission while considering different bandwidth and data quality requirements of heterogeneous end users. A packet-marking scheme is proposed that enables application-aware selective drop and forwarding of packets at intermediate overlay nodes during network congestion to further enhance the QoS received by the end users under dynamic network conditions. Effectiveness of the transport services based on application-aware architecture framework is demonstrated by the one-to-many high-bandwidth time-series radar data dissemination protocol for CASA (Collaborative Adaptive Sensing of the Atmosphere) application. Performance analysis is performed using Internet based Planetlab test bed and emulation test bed. Experiment results demonstrate that under similar network conditions and available bandwidth, application-aware processing at overlay nodes significantly improves the quality of the time-series radar data delivered to the end users compared to case when no such application-aware processing is performed. Moreover, it is shown that application-aware congestion control protocol does not degrade the performance of already existing TCP cross-traffic on the network. Scalability analysis of application-aware congestion control protocol shows that it is able to schedule data at cumulative rate of more than 700Mbps without degrading the QoS received by multiple end users. <br> Freshness of the data received by end users is an important QoS parameter for mission-critical sensor network applications. A model for tardiness of data is developed for evaluating the impact of network dynamics such as packet losses, random delay, packet reordering, and sampling rate on the freshness of the data in sensor networks. Tardiness profiles can be generated using this model for a given sensor network, which are useful for analyzing the suitability of the network infrastructure/configuration for a given mission-critical application. Alternatively, applications may use the tardiness model to adapt network operating parameters such as transmission power and sampling rate to achieve the application-specific freshness of the data. Tradeoffs between energy consumption and tardiness of the data in a wireless sensor network are also investigated. Tardiness model based result shows that it is significantly more energy efficient to achieve the desired freshness of data by adapting transmission power instead of sampling rate. It is shown that in a multi-hop wireless network there exists an optimal number of relay nodes between source and sink node that leads to minimum tardiness of data. Applications of tardiness model include (i) the estimation of error in the end results due to use of stale data in computations, and (ii) performance analysis of sensor network routing protocols by comparing tardiness of data due to selection of different paths by different routing protocols between source and sink nodes. <br>

Adviser: Prof. Anura Jayasumana
Co-Adviser: Prof. V. Chandrasekar
Non-ECE Member: Prof. Wim Bohm (CS)
Member 3: Prof. Sanjay Rajopadhye (CS)
Addional Members:


Program of Study: