The project aims to use workload forecasting in approximate query evaluation applications. For this purpose, existing forecasting techniques are leveraged to develop an approach for forecasting a sequence of workloads for a future time interval based on the data access pattern. These applications operate in two phases: off-line compression and on-line query evaluation. In the first phase, an access pattern is extracted from a sequence of accessed data elements mapped from the queries in a log, and is used to forecast a workload for each subinterval of a given future time interval. This generates a sequence of workloads. The workload information is then used to compress the data elements targeted for each subinterval. The compressed data elements are then stored on disk and, at run-time, loaded and translated into a main memory data structure needed for query evaluation in each subinterval. These steps of forecast, compression, and loading combined directly influence the trade-off between the approximate query result accuracy and query speed, thereby necessitating an optimization. The research will have immediate impacts in application areas needing the approximate query evaluation abilities of DBMSs, particularly those needing to use limited system resources effectively through workload forecasting. Furthermore, the developed techniques will have impacts in various fields of science and engineering by enabling efficient and effective utilization of limited resources based on forecasted workload.
