Combining Databases and Signal Processing in Plato

Sensors generate large amounts of spatiotemporal data that have to be stored and analyzed. However, spatiotemporal data still lack the equivalent of a DBMS that would allow their declarative analysis. We argue that the reason for this is that DBMSs have been built with the assumption that the stored data are the ground truth. This is not the case with sensor measurements, which are merely incomplete and inaccurate samples of the ground truth. Based on this observation, we present Plato; an extensible DBMS for spatiotemporal sensor data that leverages signal processing algorithms to infer from the measurements the underlying ground truth in the form of statistical models. These models are then used to answer queries over the data. By operating on the model instead of the raw data, Plato achieves significant data compression and corresponding query processing speedup. Moreover, by employing models that separate the signal from the noise, Plato produces query results of higher quality than even the original measurements.