Opportunistic Pervasive Computing with Domain-Oriented Virtual Machines

The paper targets heterogeneous sensor-actuator networks, in which nodes differ as to resources (sensors and actuators) they are equipped with. Each node contributes its specific sensors and actuators to be used by applications. The key assumption of "opportunistic pervasive computing" is that the actual mix of nodes (and that of available resources) is not known in advance to the programmer. An opportunistic pervasive computing application is supposed to take the best advantage of whatever sensors and actuators happen to be available in the network. The paper presents a technique that can be used in middleware layers supporting such applications. The technique uses virtual machines to orderly expose sensor and actuator resources of a node to the programmer. The virtual machines are domain-oriented, node specific and able to work with the resources at multiple levels of abstraction. They can be implemented on severely constrained nodes (e.g., of the TinyOS class)