Location via proxy:   [ UP ]  
[Report a bug]   [Manage cookies]                
SlideShare a Scribd company logo

Using S3 Select to Deliver 100X Performance Improvements Versus the Public Cloud

•
•
Databricks
Databricks

Using S3 Select with MinIO's object storage can provide 100x performance improvements over AWS S3. S3 Select offloads filtering of data to storage, supporting formats like CSV, JSON, and Parquet. MinIO accelerated S3 Select performance by using techniques like zero-copy parsing and SIMD to process data 10x faster. With ongoing work, S3 Select on MinIO using SIMD could achieve additional speedups versus AWS S3 Select.

1 of 15
Using S3 Select to Deliver 100X Performance Improvements Versus the Public Cloud Frank Wessels CTO, MinIO
S3 Select ▪ Recent addition to S3 API ○ Offload filtering to storage ○ Formats: CSV, JSON, Parquet ▪ Advantages ○ Faster ○ Less network traffic ○ Smaller compute nodes ￭ S3 Select for Spark ○ https://github.com/minio/spark-select Before After Up to 400% faster Up to 80% Cheaper Applications Applications S3 SELECT 2
3 MinIO is a high performance, distributed object storage server, designed for peta-scale data infrastructure. S3-Compatible Scalable PerformantSimple Optimized for Intel/ ARM/Power9 CPUs Introduction to MinIO
4 Global Scale
5 Focus on Performance
6 S3 Select Performance on AWS Format Time (s) Records Throughput csv 5.46 733K/s 94 MB/s json 14.28 280K/s 98 MB/s parquet 32.25 124K/s 4.3 MB/s
7 Evaluation (“where”) Processing (“select”) CSV JSON Parquet Parsing Parsing Loading Accelerating S3 Select on minio
8 Manage memory allocations: garbage collected vs. non-garbage collected Source: https://bitbucket.org/ewanhiggs/csv-game First 10X Acceleration: Zero Copy
9 ▪ SIMD = Single Instruction Multiple Data ○ Intel: AVX2 ▪ Process 32 bytes in parallel ○ delimiter / separator detection ○ bitmap handling & parsing ○ string compares ▪ Performance (single core) Second 10X Acceleration: SIMD
10 ▪ Same queries as before ○ minio with select-simd vs AWS S3 Results using select-simd
Demo ■ Source data ○ parking-citations.csv (25M rows / 3.5 GB) ■ AWS region ○ us-east-1 ■ minio with select-simd-integration branch running on a single instance: c5.2xlarge (8 vCPUs) ■ mc client running in same region on c5.large instance
12 ▪ Works in progress ○ Initial focus on CSV ▪ Next: add support for ○ Parquet ○ JSON: https://github.com/lemire/simdjson ▪ Investigate AVX-512 ○ erasure coding ▫ AVX-512 4x speedup over AVX2 ○ k-registers are great / 2KB on-core register space ▪ Dynamic code generation (think LLVM) Status and what’s next
Power9 CPUs PCIe Gen4 24x NVMe Dual Mellanox CX5 (4x100 GbE/s) High performance object storage 13
▪ Benefits ○ Faster queries ○ Less network traffic ○ Smaller compute needs ▪ Stay tuned for overall impact ○ S3 “plain” vs S3 Select ○ minio/simd-select vs AWS S3 Select S3 Select benefits for Spark
Questions? Visit our booth #509 @minio https://github.com/minio/minio https://slack.minio.io https://minio.io

More Related Content

Using S3 Select to Deliver 100X Performance Improvements Versus the Public Cloud

  • 1. Using S3 Select to Deliver 100X Performance Improvements Versus the Public Cloud Frank Wessels CTO, MinIO
  • 2. S3 Select ▪ Recent addition to S3 API ○ Offload filtering to storage ○ Formats: CSV, JSON, Parquet ▪ Advantages ○ Faster ○ Less network traffic ○ Smaller compute nodes ￭ S3 Select for Spark ○ https://github.com/minio/spark-select Before After Up to 400% faster Up to 80% Cheaper Applications Applications S3 SELECT 2
  • 3. 3 MinIO is a high performance, distributed object storage server, designed for peta-scale data infrastructure. S3-Compatible Scalable PerformantSimple Optimized for Intel/ ARM/Power9 CPUs Introduction to MinIO
  • 6. 6 S3 Select Performance on AWS Format Time (s) Records Throughput csv 5.46 733K/s 94 MB/s json 14.28 280K/s 98 MB/s parquet 32.25 124K/s 4.3 MB/s
  • 7. 7 Evaluation (“where”) Processing (“select”) CSV JSON Parquet Parsing Parsing Loading Accelerating S3 Select on minio
  • 8. 8 Manage memory allocations: garbage collected vs. non-garbage collected Source: https://bitbucket.org/ewanhiggs/csv-game First 10X Acceleration: Zero Copy
  • 9. 9 ▪ SIMD = Single Instruction Multiple Data ○ Intel: AVX2 ▪ Process 32 bytes in parallel ○ delimiter / separator detection ○ bitmap handling & parsing ○ string compares ▪ Performance (single core) Second 10X Acceleration: SIMD
  • 10. 10 ▪ Same queries as before ○ minio with select-simd vs AWS S3 Results using select-simd
  • 11. Demo ■ Source data ○ parking-citations.csv (25M rows / 3.5 GB) ■ AWS region ○ us-east-1 ■ minio with select-simd-integration branch running on a single instance: c5.2xlarge (8 vCPUs) ■ mc client running in same region on c5.large instance
  • 12. 12 ▪ Works in progress ○ Initial focus on CSV ▪ Next: add support for ○ Parquet ○ JSON: https://github.com/lemire/simdjson ▪ Investigate AVX-512 ○ erasure coding ▫ AVX-512 4x speedup over AVX2 ○ k-registers are great / 2KB on-core register space ▪ Dynamic code generation (think LLVM) Status and what’s next
  • 13. Power9 CPUs PCIe Gen4 24x NVMe Dual Mellanox CX5 (4x100 GbE/s) High performance object storage 13
  • 14. ▪ Benefits ○ Faster queries ○ Less network traffic ○ Smaller compute needs ▪ Stay tuned for overall impact ○ S3 “plain” vs S3 Select ○ minio/simd-select vs AWS S3 Select S3 Select benefits for Spark
  • 15. Questions? Visit our booth #509 @minio https://github.com/minio/minio https://slack.minio.io https://minio.io