Programming ad-hoc networks

Mobile ad-hoc networks of small wireless devices such as PDAs and smart phones represent a new distributed computing platform. The fact that the networks are mobile, dynamic, and embedded in the physical world poses challenges to new designs in programming models, languages, compilers, and runtime systems. For a network highly embedded in the physical world, location is crucial. A node may or may not participate in a computing task depending on where it is. Since the network has a changing topology due to node movement or failure, computation or data can not be statically assigned. Nodes need to be dynamically discovered at runtime.; In this thesis, I propose a programming model and programming language, called SpatialViews, for mobile ad-hoc networks. SpatialViews is a Java extension. The language supports location-aware service discovery, thread-level transparent parallel programming, and in-network data aggregation. The language is designed to be data race free using a type system, which is statically and dynamically checked. A prototype of SpatialViews was implemented by extending SUN JDK 1.3.1 and J2ME/MIDP 1.0. The compiler automatically parallelizes a SpatialViews program and generates mobile code migrating, replicating, and cooperating among nodes.; I will present experimental results about the performance, failure resilience, and energy efficiency based on our prototype system. And I will present applications written in SpatialViews, including sensor data collection and aggregation, computation off-loading for understanding of mobile camera images, augmented-reality games, and personalized real-time status query for mass transportation systems.