BlazingSQL

View Current Viewing Revision #8 from 12/05/2018 4:20 p.m.

BlazingSQL is a distributed GPU-accelerated SQL engine with data-lake integration. It is ACID-compliant. BlazingSQL targets ETL workloads and aims to perform efficient read IO and OLAP querying. BlazingDB refers to the company and BlazingSQL refers to the product. It is currently under active development with 15 employees. BlazingDB has offices in San Franscisco and Peru.

History

BlazingSQL started as a GPU table joiner for multi-terabyte databases. The Aramburu brothers, Rodrigo and Felipe, founded a company in 2013 that provided analytical solutions and needed to speed up joins for pension fraud detection. The system is closed-source with a free community binary. It integrates with the open-source open GPU data science initiative, RAPIDS, which relies on NVIDIA GPUs.

Query Interface

SQL

BlazingSQL exposes a Python connector for executing SQL commands.

Storage Organization

Log-structured

BlazingSQL appears to be log-structured.

Joins

Hash Join

BlazingSQL supports hash joins, e.g. on strings. It is not clear what other join types are supported.

Storage Model

Decomposition Storage Model (Columnar)

BlazingSQL is a column-store. To execute a query, it compresses and transmits relevant columns to the GPU. On the GPU, data is represented as a GPU DataFrame (GDF). GDFs are built on top of Apache Arrow, which is a columnar in-memory format.

Views

Virtual Views

BlazingSQL 1.3 supported the CREATE VIEW command. It is unclear if the views are virtual or materialized.

System Architecture

Shared-Nothing

BlazingSQL worker nodes push information to each other whenever required. There is a notion of a distributed cache, and nodes can ask each other for cached data-lake data.

Query Execution

Vectorized Model

BlazingSQL operations are vectorized on the GPU (SIMD).

Isolation Levels

Snapshot Isolation

BlazingSQL supports Snapshot Isolation, it is unclear if other options are supported.

Hardware Acceleration

GPU

BlazingSQL is hardware-accelerated with NVIDIA GPUs. Relevant columnar data is compressed, cached and sent to the GPU. The GPUs are used to speed up transforms, predicates, running predicates while skipping metadata, and to perform accelerated joins.

Indexes

Not Supported

BlazingSQL does not appear to support indexes.

Concurrency Control

Multi-version Concurrency Control (MVCC)

BlazingSQL supports snapshot isolation, which is most likely achieved with MVCC.

Compression

Dictionary Encoding Delta Encoding Run-Length Encoding Bit Packing / Mostly Encoding

BlazingSQL supports compressing and decompressing directly on the GPU. It accepts a variety of input formats such as Apache Parquet, BlazingDB Simpatico (GPU-compressed distributed files), and GDF (GPU dataframes built on Apache Arrow). Data is then sent to the GPU compressed. It is able to operate directly on compressed data.