Sensor networks are an exciting new technology that promise to revolutionize the environment we live in by creating ambient intelligent spaces. In the current software model, applications are statically loaded onto the network at compile time. This means applications cannot read to changes in the underlying network and code reuse is rare because applications are so tightly coupled with hardware. Sensor network deployments suffer from a lack of standard software APIs that allows applications to interact with the network.

This dissertation presents SNSP, a distributed service-based operating system for sensor networks. SNSP presents an integrated interface to applications, which abstracts the distributed nature of the underlying network. It enables dynamic application and content management in the sensor network. SNSP's core consists of seven OS-level services that manage content, discover network resources, monitor resource utilization, dynamically map network applications, provide fault detection and recovery, migrate applications and implement security for the sensor network. Programmers can write services that become a reusable library of SNSP code. The dissertation outlines a programming language and integrated development environment for programmers.

Further, the mechanisms for content management and replication and task allocation (mapping) are studied in more detail. Three replication schemes are compared via simulation. Results indicate that the probabilistic scheme has the best performance in terms of cost per data access and increased data availability. Moreover, three task allocation schemes are also compared. The third algorithm, a hybrid genetic search and market bidding protocol, outperforms the other two algorithms. However, due to its twelve times higher computation and 51% higher communication cost the greedy algorithm is preferred.

As a proof-of-concept, SNSP is demonstrated on a TelosB and Mica2 (with TinyOS) testbed implementation. The testbed allows applications on different hardware platforms (Mica2 and TelosB) to coexist. Measurements from the testbed indicate that the content management algorithm lowers data access cost and also the time to map a process onto the network.