Modern analytical systems are typically optimized for total runtime, but not for fast first results, even though early answers are often essential in exploratory analysis and emerging agent-driven workflows. Hence we introduce SmartRabbit. The core idea behind SmartRabbit is to run two coordinated query plans at the same time: an interactive plan that produces early, prefix-correct answers, and a latency-optimized blocking plan that completes the full query efficiently. SmartRabbit is implemented on top of Apache AsterixDB and evaluated on a multi-node cluster environment.
We try to tackle query optimization by fluid denormalization of our schema. ColumnCast analyzes query workloads and database statistics to determine candidate attributes that can be duplicated to improve query processing. This process - which we refer to as grafting can help in multiple ways: elimination of a join, predicate pushdown and even aggregation pushdown. Preliminary results show a speed up upto 1.75x with only a 8% increase in storage overhead.