ParGRES - [prototype] PostgreSQL extension for parallel query processing in shared-nothing architectures
This code devoted to demonstration of one approach [1] to parallel query execution in a shared-nothing architecture.
In accordace to the approach, parallel DBMS uses PostgreSQL core as a container for tuples. PDBMS manages metadata about data distribution. Input query is transfered to the nodes contained some part of tuples of distributed relations, involved into the query. Correctness of JOIN, Aggregate and other operations is provided by a parallel plan generator. It inserts custom exchange nodes into the plan positions, that needs tuples shuffling between nodes.
This code do not take into consideration such problems as global snapshot and distributed commit. Thereunder all transactions that need writing to distributed relations must be executed in sequental mode.
Andrey Lepikhov a.lepikhov@postgrespro.ru, Postgres Professional, Moscow, Russia
Lepikhov A.V., Sokolinsky L.B. Query Processing in a DBMS for Cluster Systems // Programming and Computer Software. 2010. Vol. 36. No. 4. P. 205-215. DOI: 10.1134/S0361768810040031