DB25

Next-Generation Hybrid Transactional/Analytical Processing (HTAP) Database System with SIMD-Optimized SQL Parser

10x

Faster Parsing

SIMD

Optimized

HTAP

Architecture

100%

Open Source

View on GitHub Read the Paper

Project Overview

What is DB25?

DB25 is an experimental database system that combines the best of transactional and analytical processing in a single, unified architecture. It features a revolutionary SIMD-optimized SQL parser that achieves 10x performance improvements over traditional parsers.

Built from the ground up with modern hardware in mind, DB25 leverages CPU vectorization, cache-aware algorithms, and lock-free data structures to deliver unprecedented performance for mixed workloads.

Key Innovations

  • ▸ SIMD SQL Parser: First-of-its-kind vectorized SQL parser using AVX-512 instructions
  • ▸ Unified Storage: Single storage engine for both OLTP and OLAP workloads
  • ▸ Adaptive Query Processing: Runtime optimization based on workload patterns
  • ▸ Zero-Copy Data Path: Minimizes data movement across system components

System Architecture

Parser Layer

SIMD-accelerated SQL parsing with parallel tokenization and syntax analysis

  • • AVX-512 vectorized tokenizer
  • • Parallel syntax validation
  • • Zero-allocation parsing
  • • Token packing optimization

Storage Engine

Hybrid row-column storage with adaptive compression and indexing

  • • Multi-version concurrency control
  • • Columnar compression
  • • Adaptive indexing
  • • Lock-free data structures

Query Processor

Adaptive query execution with runtime optimization and vectorization

  • • Vectorized execution
  • • JIT compilation
  • • Cost-based optimization
  • • Parallel query execution

Project Components

DB25 SQL Parser

Independent SIMD-optimized SQL parser achieving 10x performance over traditional parsers

C++ SIMD AVX-512
GitHub Paper

DB25 SQL Tokenizer

High-performance SQL tokenizer with parallel processing capabilities

C++ Parallel Zero-Copy
GitHub

DB25 Token Packing

Efficient token compression and packing for reduced memory footprint

C++ Compression Memory-Efficient
GitHub

DB25 Database Core

Complete HTAP database system with all components integrated

C++20 HTAP Production-Ready
GitHub Repository

Performance Benchmarks

SQL Parsing Performance

DB25 Parser 10x faster
PostgreSQL Parser Baseline
MySQL Parser 1.2x baseline

HTAP Workload Performance

Mixed OLTP/OLAP 5x throughput
Pure OLTP 3x throughput
Pure OLAP 8x throughput

Benchmark Environment: Intel Xeon Platinum 8375C (Ice Lake) • 128GB RAM • NVMe SSD • TPC-H SF100 • Mixed workload with 70% OLTP, 30% OLAP

Get Started with DB25

Clone the repository and start building with the next-generation database system

$ git clone https://github.com/space-rf-org/DB25.git
$ cd DB25
$ mkdir build && cd build
$ cmake ..
$ make -j8
$ ./db25 --help
View on GitHub Documentation