Presentation is loading. Please wait.

Presentation is loading. Please wait.

The Yin and Yang of Processing Data Warehousing Queries on GPUs

Similar presentations


Presentation on theme: "The Yin and Yang of Processing Data Warehousing Queries on GPUs"— Presentation transcript:

1 The Yin and Yang of Processing Data Warehousing Queries on GPUs
Two opposite forces that are interconnected and interdependent in natural world The Yin and Yang of Processing Data Warehousing Queries on GPUs Yuan Yuan, Rubao Lee, Xiaodong Zhang The Ohio State University 11/15/2018

2 GPUs: Powerful and Programmable
Performance (GFLOPS) After 10 years’ R&D, GPUs have evolved from dedicated graphics processors into high performance, general purpose computing devices 2002 2012 11/15/2018

3 GPUs In High Performance Computing
No. 1 Titan: 261,632 NVIDIA Cores 11/15/2018

4 GPUs: massive parallel computing units
DW workloads: rich data parallelism A decade of research efforts in database community [SIGMOD03] [VLDB04] [SIGMOD04] [SIGMOD06] [SIGMOD08] [SIGMOD10] [VLDB10] [VLDB11] [VLDB12] 11/15/2018

5 GPUs in DW Production Systems?
None! 11/15/2018

6 Why such general purpose GPUs have not been adopted for critical query processing?
Query Characteristics Software Techniques Hardware Advancement 11/15/2018

7 Query Processing on GPUs
CPU Yang Core Core Core Core GPU Core Core Core Core Device Kernel Execution Host Memory Device Memory D A T A PCIe Yin PCIe Data Transfer 11/15/2018

8 Experimental Environment
Hardware CPU Intel Core i7 3770k NVIDIA GTX 480, 580 and 680 GPU Query Engine Prototype Automatic translator from SQL to highly optimized CUDA programs (based on YSmart) Column store Workload Star Schema Benchmark 11/15/2018

9 Unbalanced Yin and Yang of SSBM
Kernel execution time varies greatly Most queries are dominated by PCIe data transfer 11/15/2018

10 Understand “Where does time go?”
How do different query characteristics affect query performance? How does GPU hardware advancement affect the performance? How do software optimizations affect the performance? 11/15/2018

11 GPU Hardware Parameters
GTX 480 GTX 580 GTX 680 Year 2010 2011 2012 Architecture Fermi Kepler # of Cores 480 512 1536 GFLOPS 1345 1581.1 3090.4 Memory BW(GB/s) 177.4 192.4 192.3 PCIe 2.0 3.0 11/15/2018

12 Limited Performance improvement by GPU Arch
480 580 680 11/15/2018

13 More Performance Improvement by PCIe Bandwidth
11/15/2018

14 Performance Prediction
10%-15% improvement 30%-35% improvement Limited benefits from near future GPU advancement 11/15/2018

15 Software Optimization
Techniques Data Compression Invisible Join Transfer Overlapping 11/15/2018

16 The Impact of Compression on SSBM
Effective: high selectivity, and less projected columns baseTransfer compressTransfer baseKernel compressKernel Ineffective: high selectivity, and with projected dim columns Ineffective: extreme low selectivity Reduced PCIe transfer time 11/15/2018

17 The Impact of Invisible Join on SSBM
Effective: more projected columns from dimension table, with high selectivity baseTransfer inviTransfer baseKernel inviKernel Effective: high Selectivity projected dim columns Ineffective: extreme low selectivity No effect on PCIe transfer 11/15/2018

18 The Impact of Transfer Overlapping on SSBM
Effective: low selectivity, more projected column from fact table Ineffective: high selectivity Effective: extreme low selectivity 11/15/2018

19 So, why GPUs not adopted in DWs?
Complicated and subtle choices of query optimization techniques. Limited usage of GPU hardware resources and unlikely benefit from GPUs advancement due to unbalanced Yin and Yang. Lack of efficient system software support for memory management and task concurrency. 11/15/2018

20 Take Home Messages Data Software Hardware
Schema design should take into account GPU features and avoid data alignment issues. Software Compression, invisible join and transfer overlapping are the most effective techniques for GPU query processing, but they favor different kind of queries Hardware Query performance are bounded by PCIe bandwidth and GPU device memory bandwidth, but limited benefits from the advancement of GPU hardware. 11/15/2018

21 Thank you! Questions? 11/15/2018


Download ppt "The Yin and Yang of Processing Data Warehousing Queries on GPUs"

Similar presentations


Ads by Google