Date of Award


Document Type

Open Access Dissertation

Degree Name

Doctor of Philosophy (PhD)

Degree Program

Computer Science

First Advisor

Deepak Ganesan

Second Advisor

Arun Venkataramani

Third Advisor

Prashant Shenoy

Subject Categories

Computer Sciences


Today many large scale sensor networks have emerged, which span many different sensing applications. Each of these sensor networks often consists of millions of sensors collecting data and supports thousands of users with diverse data needs. Between users and wireless sensors there are often a group of powerful servers that collect and process data from sensors and answer users' requests. To build such a large scale sensor network, we have to answer two fundamental research problems: i) what data to transmit from sensors to servers? ii) how to transmit the data over wireless links? Wireless sensors often can not transmit all collected data due to energy and bandwidth constraints. Therefore sensors need to decide what data to transmit to best satisfy users' data requests. Sensor network users can often tolerate some data errors, thus sensors may transmit data in lower fidelity but still satisfy users' requests. There are generally two types of requests-raw data requests and meta-data requests. To answer users' raw data requests, we propose a model-driven data collection approach, PRESTO. PRESTO splits intelligence between sensors and servers, i.e., resource-rich servers perform expensive model training and resource-poor sensors perform simple model evaluation. PRESTO can significantly reduce data to be transmitted without sacrificing service quality. To answer users' meta-data request, we propose a utility-driven multi-user data sharing approach, MUDS. MUDS uses utility function to unify diverse meta-data metrics. Sensors estimate utility value of each data packet and sends packets with highest utility first to improve overall system utility. After deciding what data to transmit from sensors to servers, the next question is how to transmit these data over wireless links. Wireless transport often suffers low bandwidth and unstable connectivity. In order to improve wireless transport, I propose a clean-slate re-design of wireless transport, Hop. Hop uses reliable perhop block transfer as a building block and builds all other components including hidden-terminal avoidance, congestion avoidance, and end-to-end reliability on top of it. Hop is built based on three key ideas: a) hop-by-hop transfer adapts to the lossy and highly variable nature of wireless channel significantly better than end-to-end transfer, b) the use of blocks as the unit of control is more efficient over wireless links than the use of packets, and c) the duplicated functionality in different layers in the network stack should be removed to simplify the protocol and avoid complex interaction.