Systems and methods for artificial intelligence-based data system optimization
View Patent ↗In some aspects, the disclosure is directed to methods and systems for access path selection optimization for relational database queries. Machine learning systems, such as neural networks, may be used to determine costs for each path. A neural network may be trained based on cost determinations from scans and indexes of the tree for an initial set of queries, and then may be used to predict costs for additional queries such that a path may be selected. The training data may be periodically refreshed or updated, or may be refreshed or updated responsive to changes in hardware or computing environment.
1 . A method for training an optimizer for a data system, comprising:
for each of a plurality of queries of a database maintained by one or more computing devices, each query associated with one or more features:
receiving, by a computing system, the query,
executing, by the computing system, the query using a first query methodology,
measuring, by the computing system, one or more characteristics of the one or more computing devices while executing the query using the first query methodology,
generating, by the computing system, a first tuple comprising the one or more features of the query and the one or more characteristics measured while executing the query using the first query methodology,
executing, by the computing system, the query using a second query methodology,
measuring, by the computing system, the one or more characteristics of the one or more computing devices while executing the query using the second query methodology, and
generating, by the computing system, a second tuple comprising the one or more features of the query and the one or more characteristics measured while executing the query using the second query methodology; and
training, by the computing system, a learned optimizer from the first and second tuple generated for each of the plurality of queries, the learned optimizer comprising a neural network trained to estimate a cost for performing a new query via each of the first query methodology and the second query methodology based on the one or more features of the new query.