Adaptation to Query Dynamics

Just as sensor data exhibits time-varying behavior, query patterns can also change over time. In particular, the query tolerance demanded by queries may change over time, resulting in more or fewer data pulls. The proxy can adapt the value of the threshold parameter $\delta$ in Equation 6 to directly influence the fraction of queries that trigger data pulls from remote sensors. If the threshold $\delta$ is large relative to the mean error tolerance of queries, then the number of pushes from the sensor is small and the number of pulls triggered by queries is larger. If $\delta$ is small relative to the query error tolerance, then there will be many wasteful pushes and fewer pulls (since the cached data is more precise than is necessary to answer the majority of queries). A careful selection of the threshold parameter $\delta$ allows a proxy to balance the number of pushes and the number of pulls for each sensor.

To handle such query dynamics, the PRESTO proxy uses a moving window average to track the mean error tolerance of queries posed on the sensor data. If the error tolerance changes by more than a pre-defined threshold, the proxy computes a new $\delta$ and transmits it to the sensor so that it can adapt to the new query pattern.