Microsoft Research333 тыс
Опубликовано 24 октября 2019, 20:14
Selecting an appropriate set of indexes to build in a database for a given workload can result in significant reductions in query execution cost, e.g., CPU time. Being able to fully automate index recommendation and implementation is a significant value-add for a cloud database service provider. One key requirement of automated index implementation for production systems is that creating or dropping indexes does not cause significant query performance regressions. Such regressions, where a query’s execution cost increases after changing the indexes, is a major impediment to fully-automated indexing as users desire to enforce a no query regression constraint.
In this talk, Bailu Ding gives an overview of the architecture of auto-indexing. She then covers two research directions in this area from her previous work: Plan Forcing for regression recovery and AI Meets AI for regression prevention.
In this talk, Bailu Ding gives an overview of the architecture of auto-indexing. She then covers two research directions in this area from her previous work: Plan Forcing for regression recovery and AI Meets AI for regression prevention.