Search

The need for interactive ad-hoc analytics has been steadily rising. The traditional batch processing of large data takes order of seconds to several minutes for query execution, which cannot guarantee an interactive experience. With ever increasing size of data, gaining an interactive experience is non-trivial. Online aggregation solves part of the problem by returning estimated results at regular intervals. An online aggregation system allows the user to observe the progress of the query but falls short when it comes to controlling the query execution. We propose an online aggregation system that takes query priority as a dynamic user defined input, allowing the user to control the progress of query execution by adjusting the priority at runtime. The framework derives concepts from online aggregation and shared scan query execution model, coupled with our priority based scheduling to allow for dynamic user defined query reprioritization. We also show how the proposed system behaves for different underlying storage formats.