Scribd
Upload
3K views

GPU Accelerated Databases, Speeding Up Database Time Series Analysis Using OpenCL

This document discusses using GPU acceleration with databases. It outlines using OpenCL to accelerate PostgreSQL with new data types like vectors and images. Challenges include problem sizes larger than GPU RAM, setup versus runtime issues, and data transfer speeds. Example use cases presented are time series analysis, 3D content management, bioinformatics, and accelerating internal database operations. The speaker has 35 years of software experience and the goal is to make GPU acceleration easier to use and scale to larger data sizes.

Uploaded by

3dmashup
Copyright
© Attribution Non-Commercial (BY-NC)
Available Formats
Download as PDF, TXT or read online on Scribd
3K views

GPU Accelerated Databases, Speeding Up Database Time Series Analysis Using OpenCL

This document discusses using GPU acceleration with databases. It outlines using OpenCL to accelerate PostgreSQL with new data types like vectors and images. Challenges include problem sizes larger than GPU RAM, setup versus runtime issues, and data transfer speeds. Example use cases presented are time series analysis, 3D content management, bioinformatics, and accelerating internal database operations. The speaker has 35 years of software experience and the goal is to make GPU acceleration easier to use and scale to larger data sizes.

Uploaded by

3dmashup
Copyright
© Attribution Non-Commercial (BY-NC)
Available Formats
Download as PDF, TXT or read online on Scribd
You are on page 1/ 29

GPU Accelerated Databases

Database Driven OpenCL Programming


Tim Child 3DMashUp CEO

Outline
Speakers Biography Outline Solution Goals OpenCL Programming Challenge Review of GPU Accelerated Databases Swiss Army Knife of Data OpenCL Bindings to PostgreSQL Challenges Example Use Cases Benefits of the Approach Q&A

Speakers Bio
Tim Child 35 years experience of software development Formerly
VP Engineering, Oracle Corporation VP Engineering, BEA Systems Inc. VP Engineering , Informix Leader at Illustra, Autodesk, Navteq, Intuit,

30+ years experience in 3D, CAD, GIS and DBMS

Goals
Develop New Applications
Develop new GPU Accelerated Database Applications that are computationally intensive.

Ease of Use
Make use GPU accelerated code easier to use Make GPU accelerated code more mainstream to Information Technology

Data Scalability
Scale GPU application data size

Enhance existing database internal operations

OpenCL Programming Challenge


Write an OpenCL Application that : Reads data from DBMS or File Publishes Results as Web Pages Handles Frequent Data Updates Data Size >> System RAM >> GPU RAM

Possible Solutions
Other Choices ??

or C/C++ Binding using Web CGI Database Driven Java/Perl/Python Bindings in App Server GPU Programming

REVIEW OF GPU ACCELERATED DATABASE ARCHITECTURES

GPU Co-Process
TCP/IP DBMS Client DBMS Server

IPC / RPC
GPU Language Co-Process

GPGPU DRAM
PCI Bus

Examples 2004 Bandi, Sun, et al Many others

Data Tables

GPGPU

GPU Hosted Data Architecture


PCI Bus TCP/IP DBMS Client DBMS Sever + GPU Host Data Indices Copy GPGPU DRAM Data Tables Copy

GPGPU

Examples 2008 Bakkum, Skardon 2010 Palo OLAP 2010 ParStream 2011 Kaczmarski

Data Tables

Procedural Language Architecture


TCP/IP PCI Bus DBMS Server Results GPGPU Host Queries

GPGPU DRAM

DBMS Client

10G B

RAM Cache

GPGPU

Examples 1995 Illustra/Intel 2010 3DMashUp Data Tables

10T B

PostgeSQL Swiss Army Knife of Data


SQL Extensible Types
(Declarative Language, Set Operations)

Extensible Procedural Languages


(Java, Perl, )

Rules System

Extensible Indices

Open Source

Vibrant Community

Native APIs

Remote Data Access

PostGIS
(Vector, Raster)

OpenCL

SQL OpenCL Types


Vector Types
cl_charX cl_ucharX cl_shortX cl_ushortX cl_floatX cl_doubleX
SQL Syntax Create table opencltypes ( id serial, matrix cl_double4[4], image image2d ); Insert into opencltypes ( matrix) values ( { 1,0,0,0, 0,1,0,0, 0,0,1,0, 0,0,0,1 } );

Images Types
image2d_t Image3d_t

Database Driven OpenCL


PostgreSQL Sever
HTTP PgOpenCL PgOpenCL SQL SQL Procedure Procedure PCIe x2 Bus TCP/IP

Web Browser

Web Server

SQL Statement

App Server

PostgreSQL GPGPU
TCP/IP

Disk I/O PostgreSQL

Data Tables

Client

OpenCL SQL Language Bindings


CREATE or REPLACE FUNCTION VectorAdd(IN Id int[], IN a real[], IN B real[], OUT C real[] ) AS $BODY$

__kernel void VectorAdd( __global int * id, __global float *a, __global float *b, __global float *c) { int i = get_global_id(0); /* Query OpenCL for the Array Subscript **/ c[i] = a[i] + b[i]; }

$BODY$
Language PgOpenCL; Select VectorAadd(Id, a, c) from Vectors;

Comparison Table

Database Driven OpenCL


Table A B Select Table to Array 100s - 1000s of Threads (Kernels)

xPU

VectorAdd(A, B) Returns C

Copy Copy

Unnest Array To Table

Table

MORE DATA TYPES

PgOpenCL Time Series Type

CL_UNSIGNED_INT, CL_INTENSITY

CL_FLOAT, CL_INTENSITY

Time Series Data


34 Years IBM data in 3NF = 8734 records Date 3/11/2003 3/10/2003 3/7/2003 3/6/2003 3/5/2003 3/4/2003 3/3/2003 Open 75.82 77.45 75.71 77 76.7 77.6 78.9 High 76.33 77.45 77.99 77.78 77.73 77.75 79 Low 75.2 75.5 75.71 76.7 76.25 76.53 77.12 Close Volume 75.35 8119200 75.7 6641300 77.9 8129200 77.07 5876300 77.73 6658000 76.7 5672200 77.3 661830

As Time Series = 34 Records, 6 Series Columns (~256 Values/Series)

Time Series Properties


Hurst Exponent Based on Fractal Dimension
0.5 Random < 0.5 Seasonal Variations > 0.5 Trending

Pearson Match Correlation Coefficient Correlation between two Time Series 1 Linear Relation Between Samples -1 Inverse Linear Relation Between Samples 0.0 No Linear Relationship between samples

FURTHER USES OF GPU ACCELERATED DBMS

Example Use Cases


GPU Accelerated Time Series 3D Content Management / GIS
Spatial Selections Coordinate Transformations Image Processing

Bioinformatics
DNA & Protein Sequence Matching

Database Internal Operations


Joins Sorting Query Planning

Example Screen 1

Example Screen 2

Example Screen 3

Example Screen 4

Example Screen 5

Type Mapping

Challenges
Problem Size
DBMS Table Size >> GPU RAM

Setup > Runtime

Extended SQL Types OpenCL Vectors Types OpenCL Image Types Time Series
Caching kernel info CPU GPU Still present SQL Queries

# Work Groups / # Work Items


Dynamic Parallelism

Runtime Partitioning
Dynamic Simplified Return Types

Data Transfer

Device Management
CPU vs. GPU
Runtime Selection

Concurrency
No Pre-emptive Multi-Tasking
Time-out Long Queries Partitioning / Scheduling

+ Overhead ( < 4s )

Map Array
Bulk Data Loaders
New Task

Summary
OpenCL

PostgreSQL

Open Source Release

Database Internal Operations

Q&A
PgOpenCL Twitter @3DMashUp Blog www.scribd.com/3dmashup OpenCL

www.khronos.org/opencl/ www.amd.com/us/products/technologies/stream-technology/opencl/ http://software.intel.com/en-us/articles/intel-opencl-sdk http://www.nvidia.com/object/cuda_opencl_new.html

You might also like