Flexs – A Logical Model for Physical Data Layout

摘要

Driven by novel application domains and hardware trends database research and development set off to many novel and specialized architectures. Particularly in the area of physical data layout, specialized solutions have shown exceptional performance for specific applications. This trend is great for research and development and for those in need of top-level performance first and foremost. For those with moderate performance needs, however, a universal but flexible database system has the benefit of lower TCO. Regarding physical data layout, the more general systems are fairly inflexible compared to the variety of physical data layouts available in specialized systems. Particularly, the macroscopic characteristics, i.e., how the data is grouped and clustered, are generally hard-coded and cannot be changed by configuration. We present Flexs, a declarative storage description language for the macroscopic characteristics of physical data layouts. Flexs allows describing physical data layouts ranging from the row and column store layouts to data layouts for irregular data such as vertical schema. Using Flexs, a storage engine can be configured to use a specific physical data layout. Flexs contributes to make specialized physical data layouts available to the broad majority of universal database systems.