Abstract
INUM is a what-if optimization technique that efficiently estimates the cost of optimal query plans under hypothetical index configurations and can thus serve as a fast alternative to conventional what-if optimization. In this paper we introduce three crucial enhancements to INUM: a principled method to handle query plans with Nested-Loop Join (NLJ) operators (to improve estimation accuracy); a method to reduce the time to preprocess a query in the workload (to reduce setup latency); and, a method to prune the amount of information stored per query (to improve estimation efficiency). We demonstrate experimentally that these improvements make INUM 5x faster and improve median estimation accuracy by 79%. Our work extends significantly the scope of workloads and tuning problems to which INUM can be applied.