Comment 8. Other datasets on which ARIMA works. Other models which can be used in PRESTO.
This paper described PRESTO, a model-driven predictive data management architecture for hierarchical sensor networks. In contrast to existing techniques, our work makes intelligent use of proxy and sensor resources to balance the needs for low-latency, interactive querying from users with the energy optimization needs of the resource-constrained sensors. A novel aspect of our work is the extensive use of an asymmetric prediction technique, Seasonal ARIMA [1], that uses proxy resources for complex model parameter estimation, but requires only limited resources at the sensor for model checking. Our experiments showed that PRESTO yields an order of magnitude improvement in the energy required for data and query management, simultaneously building a more accurate model than other existing techniques. Also, PRESTO keeps the query latency within 3-5 seconds, even at high query rates, by intelligently exploiting the use of anticipatory pushes from sensors to build models, and explicit pulls from sensors. Finally, PRESTO adapts to changing query and data requirements by modeling query and data parameters, and providing periodic feedback to sensors. As part of future work, we plan to (i) extend our current models to other weather phenomena beyond temperature and to other domains such as traffic and activity monitoring, and (ii) design spatio-temporal models that exploit both spatial and temporal correlations between sensors to further reduce communication costs.