Learning How to Optimize Data Access in Polystores

摘要

Polystores provide a loosely coupled integration of heterogeneous data sources based on the direct access, with the local language, to each storage engine for exploiting its distinctive features. In this framework, given the absence of a global schema, a common set of operators, and a unified data profile repository, it is hard to design efficient query optimizers. Recently, we have proposed QUEPA, a polystore system supporting query augmentation, a data access operator based on the automatic enrichment of the answer to a local query with related data in the rest of the polystore. This operator provides a lightweight mechanism for data integration and allows the use of the original query languages avoiding any query translation. However, since in a polystore we usually do not have access to the parameters used by query optimizers of the underlying datastores, the definition of an optimal query execution plan is a hard task, as traditional cost-based methods for query optimization cannot be used. For this reason, in the effort of building QUEPA, we have adopted a machine learning technique to optimize the way in which query augmentation is implemented at run-time. In this paper, after recalling the main features of QUEPA and of its architecture, we describe our approach to query optimization and highlight its effectiveness.