The Technology
Large datasets play a vital role, particularly in modern AI applications. To ensure efficient query processing and minimize disk access, advanced query optimization capabilities are necessary.
The containment rate measures the percentage of result tuples from query Q1 that are also present in the result of query Q2, based on a specific database. We propose an innovative approach called Containment Rate Network (CRN), which employs specialized deep learning techniques to directly estimate containment rates between query pairs within a given database. This estimation is crucial for accurate result-cardinality estimation, a key aspect of query optimization.
Our technique leverages estimated containment rates among queries to estimate their result-cardinalities. This estimation method can utilize CRN or incorporate existing cardinality estimation approaches without modification. In experimental evaluations on a challenging real-world database, our novel approach demonstrates significant improvements over state-of-the-art cardinality estimation methods.
Query optimization heavily relies on cardinality estimation, especially for SQL queries involving operators like AND, OR, and NOT. Additionally, queries with DISTINCT require set-theoretic cardinalities (without duplicates) for planning purposes, such as sorting options. However, many cardinality estimation methods are limited to conjunctive queries with duplicate counting, overlooking the importance of estimating cardinalities in the presence of DISTINCT, AND, OR, and NOT operators. We extend our cardinality estimation techniques to encompass this broad and practical class of queries.
Advantages
- Improvement of database access efficiency
- Estimate cardinalities of queries with the DISTINCT keyword
Applications and Opportunities
- Improvement of query plan
Business Development Contacts