3
GPU Database Acceleration on PowerEdge R940xa
Table of contents
Revisions
.............................................................................................................................................................................
2
Acknowledgements
.............................................................................................................................................................
2
Executive summary
.............................................................................................................................................................
4
1
Evolution of databases
.................................................................................................................................................
5
2
What is GPU acceleration and how does it apply to databases?
................................................................................
6
2.1
Why GPU?
..........................................................................................................................................................
6
2.2
What database operations can run on GPU?
.....................................................................................................
7
2.3
How do GPUs accelerate analytic workloads?
...................................................................................................
7
2.4
What are some examples?
.................................................................................................................................
7
2.5
How can GPU database acceleration help other workloads like Machine Learning and Deep Learning?
........
7
2.5.1
How does Brytlyt help with Machine Learning?
..................................................................................................
8
2.6
Why keep CPU: GPU ratio of 1:1 in R940xa?
..................................................................................................
10
2.6.1
Disk-IO bottleneck
............................................................................................................................................
11
2.6.2
PCIe bottleneck
................................................................................................................................................
11
3
The Dell EMC PowerEdge R940xa server
.................................................................................................................
12
3.1
CPU: GPU ratio
................................................................................................................................................
12
4
The challenge with GPU Databases
..........................................................................................................................
13
4.1
Overview of Brytlyt
............................................................................................................................................
13
4.2
Background on TPC-H Benchmarking
[1]
..........................................................................................................
14
4.3
Importance of TPC-H for Ad-Hoc Business Decision Making Activities
...........................................................
14
5
Brytlyt + Dell EMC PowerEdge R940xa Benchmarking
.............................................................................................
16
5.1
TPC-H benchmarking on PowerEdge R940xa with Brytlyt
..............................................................................
16
5.2
NY TAXI Benchmarking on PowerEdge R940xa with Brytlyt
...........................................................................
16
5.3
PowerEdge R940xa NY Taxi data runtimes
.....................................................................................................
16
5.4
Previous Benchmark
........................................................................................................................................
16
5.5
Reducing Time-to-Value for Data Analysts
......................................................................................................
16
5.6
Use Cases for Brytlyt’s GPU Database
............................................................................................................
17
6
References
.................................................................................................................................................................
18