Related Work

Sensor networks [11] have recently emerged as a promising platform for a myriad of distributed embedded applications in defense [42] and scientific exploration [21]. A typical sensor network is highly distributed and composed of thousands to hundreds of thousands of individual nodes. Communication protocols are therefore a very important research topic in sensor networks.

Most prior work on communication in sensor networks has focused on the lower layers in the protocol stack. For example, [41,45] propose MAC layer protocols designed for sensor networks. At the network layer, protocols such as DSDV [27], DSR [14], AODV [29] and TORA [25] have gained popularity as routing solutions for ad hoc wireless networks. These protocols are designed for networks with identifier-based node addressing. Recent sensor network research suggests alternative addressing schemes that do not rely on having destinations with specific identities. Instead, it has been proposed that routing in sensor networks be attribute-based where the destination is reached by its attributes such as location or sensor measurements. For example, LAR [17] and DREAM [3] propose location-aware routing protocols, where the destination is implicitly defined by its physical location. Directed diffusion [13] and the intentional naming system [1] provide routing and addressing based on data interests. A related effort is attribute-based naming [32], proposed for an Internet environment, which allows queries to be routed depending on the requested content rather than on the identity of the target machine. Our work falls in the general category of attribute-based communication. We provide an infrastructure where communication end-points are placed at the locations of specific events in the environment. Unlike prior work on attribute-based addressing, we focus on protocol dynamics that arise due to the motion of such events in the external world. We aim to maintain the persistence and uniqueness of these communication end-points as the event moves and discuss factors that affect the maximum trackable speed of these events. Also, unlike routing-layer approaches, our architecture sits in the transport layer on top of geographic forwarding.

Mobile end-points have been addressed in traditional and ad hoc computer networks. For example, Energy-aware routing protocols were proposed such as Span [6] and GAF [43] for communication between mobile nodes. In Mobile IP [26], mobile hosts are free to migrate between LAN's while they remain connected by a home agent residing at their home address. Another mechanism for maintaining mobile connections uses DNS to provide the indirection necessary to support mobility [34]. We address mobility in a different sense in our protocol. Whereas the aforementioned protocols assume moving nodes and provide for communication between these moving nodes, we assume static nodes but moving events (and thus entities) in our system and provide a migratory end-point infrastructure for communication between these moving entities. A recent protocol, called TTDD [44], addresses communication between moving sources and sinks in a sensor network. Our work differs from this in the fact that we provide for the creation and maintenance of abstract entities to facilitate communication between moving events in the network. Also, our communication end-points are bi-directional as opposed to being statically designated as sources or sinks.

Several algorithms exist that provide clustering and various granular levels of group formation on both the network and application layers. The ($\alpha$, t) framework [22,37], and Landmark Hierarchy [39] organize nodes into hierarchical groups as a solution to routing. LEACH [10], ASCENT [5], SPAN [6], and GAF [43] form groups or share data locally to conserve energy and power down unused or unneeded nodes. The AC Hierarchy [7,8,38] forms hierarchical clusters covering the entire sensor network and provides a high level programming abstraction for division or simplification of the sensor network. Finally, GLS [19] and MASH [33] provide cluster-based location or query services for locating data or nodes. Unlike these cluster-based or group-based algorithms we provide the abstraction of tracking groups linked directly to environmental events of interest. Although some of these algorithms such as GLS could be used in parallel with our work for object lookup, and although various ideas from these protocols are similar or could be used to enhance the efficiency or functionality of our modules, none provide sufficient support for entity formation around environmental events of interest and end-to-end connection establishment and maintenance between moving entities.

Our work is also complementary to several research efforts that aim to provide new abstractions and paradigms for distributed computing in sensor networks. For example, MagnetOS [2], exports the illusion of a single Java virtual machine on top of a distributed sensor network. The application programmer writes a single Java program. The run-time system is responsible for code partitioning, placement, and automatic migration such that total energy consumption is minimized. Mate [18] is another example of a virtual machine developed for sensor networks. It implements its own bytecode interpreter, built on top of TinyOS [11]. The interpreter provides high-level instructions (such as an atomic message send) which the machine can interpret and execute.

To the authors' knowledge, the communication architecture proposed in this paper is the first that provides middleware support to tracking applications for group formation around environmental events, end-to-end connection establishment between different entities, and abstract entity maintenance to ensure that a single entity is formed and maintained for every event in the environment. The architecture ensures a one to one relationship between abstract entities and environmental events thus simplifying communication. It reduces programming complexity by allowing communication with entities rather than individual nodes.