RETRI in other contexts

The idea of using RETRI, exemplified in our address-free fragmentation, has applications in other areas of a distributed sensor network. These applications all have in common a need to reference some state that has meaning over some time period and in some location. Each application's transaction density is affected by the temporal and spatial extent over which this state must remain valid. The definition of a transaction and the method of detection of collisions are both highly application-dependent.

A number of other examples are seen in sensor networks, described in more detail in [6], including:

• Interest reinforcement. Nodes that periodically transmit their sensor readings may wish to transmit more or less frequently depending on the network's level of interest in their readings. When a node transmits a sensor reading, its neighbors periodically send feedback to the transmitter indicating their level of interest. With unique addresses assigned to each transmitter, the feedback might take the form of a message such as ``Sensor #27.201.3.97, send more of your data.'' An address is not actually needed in this context; it is simply used as a way of describing data that was previously received. RETRI can serve this purpose equally well: ``Whoever just sent data with Identifier 4, send more of that.''

• Attribute-based name compression. The attributes and associated values might be quite large, but the same attribute/value pairs might be used frequently by a node. This problem has traditionally been solved by creation of a ``codebook'' mapping small identifiers to long lists of attributes. Nodes using codebooks can choose RETRI identifiers instead of traditional alternatives (using large, guaranteed-unique identifiers, or expending energy to ensure that the codes are conflict-free).

Although identifier conflicts can lead to losses or unexpected behavior, robustness to these types of errors must already be fundamental to the design of these systems, where errors are the norm due to factors such as node dynamics, changes in the environment, and the vagaries of RF connectivity. Occasional identifier collisions will have a small marginal effect, and persistent or systematic collisions are avoided by picking a new identifier for each new transaction.