?
PosDB: A Distributed Column-Store Engine
In this paper we present a novel disk-based distributed column-store, describe its architecture and discuss a number of technical solutions. Our system is essentially a query engine which was written completely from scratch. It is aimed for shared-nothing environments and supports different forms of parallel query processing.
Query processing in PosDB is organized according to the classic Volcano pull-based model which is adapted for the column-store case. Currently, we support late materialization only, and therefore employ a join index data structure to represent positional information. In our system query plan can consist of both positional and value operators. PosDB has about a dozen of core operators among which several variants of selections and joins, aggregation. We also have several operators that ensure intra-query parallelism and operators for network interoperability. In its current state the system is fully capable of processing the Star Schema Benchmark in a local and distributed environment.