Machine learning based query optimization for federated databases

A computer program product, comprising a non-transitory machine-readable storage medium having encoded therein executable code of one or more software programs, wherein the one or more software programs when executed by at least one processing device perform the following steps:

• obtaining at least one query to be processed in a federated database;
• generating at least one predictive data movement instruction to move data to a target data source when the target data source satisfies one or more of a predefined efficiency criterion with respect to a query type of said at least one query and a predefined capacity criteria at an expected execution time of said at least one query;
• and
• generating, using at least one processing device, a query execution plan for said at least one query by calculating a cost of execution for a plurality of potential target data sources and selecting a target data source for said at least one query based on said calculated cost of execution, wherein the cost of execution comprises a predicted likelihood of the target data source being busy based on one or more of scheduler load information and historic load trends.

An apparatus, comprising:

• a memory;
• and
• at least one processing device, coupled to the memory, operative to implement the following steps:
• obtaining at least one query to be processed in a federated database;
• generating at least one predictive data movement instruction to move data to a target data source when the target data source satisfies one or more of a predefined efficiency criterion with respect to a query type of said at least one query and a predefined capacity criteria at an expected execution time of said at least one query;
• and
• generating, using at least one processing device, a query execution plan for said at least one query by calculating a cost of execution for a plurality of potential target data sources and selecting a target data source for said at least one query based on said calculated cost of execution, wherein the cost of execution comprises a predicted likelihood of the target data source being busy based on one or more of scheduler load information and historic load trends.