SQream

A GPU-based Database Optimized for Big Data Analysis

SQream is a big data analysis platform as well as a GPGPU(General Purpose GPU)-based database optimized for data processing. SQream has thousands of cores that are used for the parallel processing of large datasets, delivering greater performance compared to other solutions. Its configuration and reliability have been verified on Samsung Cloud Platform, and a web-based console makes usage and installation easy.

Overview

  • Query Performance & Compatibility

    SQream DB uses the ANSI 92-compliant standard SQL query without any changes and automatically converts it to a relational query for algebra. Therefore, a GPU core can effectively handle massive parallel calculations based on the optimized query.

  • Optimal for Big Data Analytics

    Being a column-oriented database optimized for big data analysis, SQream can effectively support on-line analytical processing(OLAP). This makes it ideal for statistical analysis such as aggregate calculation by day or by account. SQream also supports data compression using a compression algorithm that aggregates similar data types and stores them.

  • GPU-based Architecture

    SQream stores data in the form of a column-by-column page called Chunk, and scans and processes data by Chunk. By assigning metadata to Chunk, data processing and
    I/O time are reduced significantly, enabling efficient processing of GPU memory transactions.

  • Individual Scalability

    SQream DB supports scaling of independent components for storage, compute node, and GPU. For instance, users can additionally scale storage during data increase.

01

04

Service Architecture

  • (Data Sources, ETL) ↔ SQream DB(Samsung Cloud Platform) → Data Science & BI
  • Data Sources: Spark, Storm, CSV, Kafka, Parquet
  • ETL: Talent, Informatica
  • SQream DB: ODBC, JDBC, Python, C++, SQL, Java, Node, JS
  • Samsung Cloud Platform: SQRERM ← GPU
  • Data Science & BI: +ableau, Power BI, Spottire, Jupyter, Qlik, H2Q.ai, SAS Viya, Apache Zeppelin, TensorFlow, R

Key Features

  • High performance data processing technology

    - Massively parallel processing technology using NVIDIA CUDA framework
    - Best architecture for large-volume statistical analysis processing : Compression algorithm processing and data classification using Group By (COUNT, SUM, AVG, MAX, MIN, etc.)
    - Parallel execute engine support concurrent query processing without decompression process

  • Cluster with infinite scalability

    - Workloads are automatically assigned to vacant GPU using SQream’s own load balancer
    - GPU can be split up to 8 pieces, maximizing concurrency

  • DB creation and management

    - Auto-provisioning (for single), lifecycle management
    - Choose VM based on the required spec
    - Provide additional storage for connection other than the OS disk
    - Provide subnet/IP, NAT IP, Security Group connection setting
    ※ Provide a separate manual for management features such as HA configuration, monitoring, and backup/restoration (User configuration item)

Pricing

    • Billing
    • Charged for VM usage
    • Being a commercial software, SQream is provided at a separate charge
    • Default option and user-added option
    • Default option : OS software, OS installation disk (100GB), 1 public IP
    • Additional charges based on user-added product/option (e.g. reserved IP)
Let’s talk

Whether you’re looking for a specific business solution or just need some questions answered, we’re here to help

Get in Touch
Share KakaoTalk Facebook Twitter Linkedin Close Share Box