PADB’s exceptional query performance begins with the Omne (pronounced “ahm-nee”) Optimizer, a new advanced parallel query optimizer built to handle complex queries within the context of PADB’s unique architecture. Specifically, the optimizer is both cost-based and rule-based; has advanced view folding, column pruning, native correlated subquery decorrelation, and query rewrite capability; and is MPP, columnar, and compression aware. It can perform join order optimization for queries with an unlimited number of joins (tested to 1000-way), and evaluates viable alternatives to provide consistently efficient query plans even for scenarios such as outer joins, and correlated sub-queries.
Most optimizers struggle with many-way joins and intermediate result sets – classic analytic query problems. As a competitive advantage, ParAccel has invested in a robust, world-class, MPP query optimizer because it makes data more accessible.
Complex Join Processing
The need to join large sets of data in complex ways is a defining characteristic of an analytic query. Therefore, efficient complex join processing is a defining characteristic of an analytic DBMS. Rather than be join averse (e.g., rely on the existence of a restricted schema or tuning structures, both of which shift the burden from the DBMS to the customer’s designers and ETL processes), PADB’s Omne Optimizer includes a patent-pending design to make good algorithmic choices about how to perform joins most efficiently.
Additionally, the ParAccel Communication Fabric also facilitates good join processing because PABD can rely on its ability to quickly distribute or broadcast join data at query time. Without a robust communication fabric, a parallel DBMS must rely on extensive data replication in order to avoid a broadcast or redistribution.
Through engineering, PADB builds into the DBMS platform elegant solutions for tough join processing issues. This allows for ongoing scalability and performance without defaulting to artificial structures, data replication or denormalization.
Query Concurrency
PADB is designed to maximize overall throughput. The design philosophy is to give each query unfettered access to the resources it needs so it may run as quickly as possible. The number of concurrent queries is configurable, but, conceptually, the approach is to allow many sessions and to throttle the query flow so the system resources are kept fully busy without creating inefficiencies (e.g., swapping) that may limit overall throughput.