Speeding Up Database Time Series Data Analysis using OpenCL

Talk Type: 
45 Minute Talk
Emerging technologies
Technical Level: 
Creative Commons - No Commercial

Compute bound PostgreSQL Data Analytics and Data Mining queries can exhibit poor performance; as PostgreSQLs’ “per process model” constrains a query to a single CPU Core. With PgOpenCL Time Series, a new data type, compute intensive queries are parallelized to take advantage of all CPU cores. If a discrete GPU, or AMD Fusion GPU, is available, queries can take advantage of the additional 100’s to 1000’s of cores for further speed ups.

The Time Series data types are specialized data types that represent a sequence of data points measured typically at successive times spaced at uniform time intervals. Some typical examples are Wall Street closing prices, climate data and digitized signals. The PgOpenCL Time Series data types utilize native OpenCL data types to take maximum advantage of GPU hardware.

PgOpenCL is a new PostgeSQL Stored Procedure language that implements Open Computer Language (OpenCL) for Parallel Processing on Multi-Core and Many-Core systems. OpenCL is freely available on Windows and Linux systems.