Abstract

ElVoidDB is a lightweight C++17 database engine designed for educational use and simple key–value workloads. It shows the basic parts of a database: storage, buffer pool, concurrency, and SQL parsing.

Why I Built This

This database is mainly inspired by my CSE 562 Database Systems class and by TacoDB. My goal is to build a real working database from scratch. I built it step by step as I learned the class concepts and saw how TacoDB is structured.

Project Goal

• Teach database internals with a real example
• Support basic SQL: SELECT, INSERT, UPDATE, DELETE
• Use a slotted-page design with 4 KB pages
• Implement an LRU buffer pool and async I/O

Architecture

The system has these layers:

1. Parser & Planner – turns SQL text into an AST
2. Execution Engine – runs commands from the AST
3. Storage Manager – reads/writes 4 KB pages
4. Buffer Pool – caches pages and flushes them in the background
5. Concurrency & Logging – simple async I/O threads

Folder Structure

ElVoidDB/

  ├─ include/

  │  ├─ Exceptions.hpp

  │  ├─ Page.hpp

  │  ├─ Storage.hpp

  │  ├─ BufferPool.hpp

  │  ├─ ThreadPool.hpp

  │  ├─ BackgroundFlush.hpp

  │  ├─ Parser.hpp

  │  └─ Commands.hpp

  └─ src/

     ├─ main.cpp

     ├─ Exceptions.cpp

     ├─ Page.cpp

     ├─ Storage.cpp

     ├─ BufferPool.cpp

     ├─ ThreadPool.cpp

     ├─ BackgroundFlush.cpp

     ├─ Parser.cpp

     └─ Commands.cpp

Build & Run (Raspberry Pi 5, 8 GB)

API Reference

• Storage::readPage(page_id) – loads a page from disk
• BufferPool::fetchPage(page_id) – gets a page with LRU policy
• Parser::parse(sql) – creates an AST from SQL
• Command::execute() – runs a SQL command

Performance Metrics

On a Raspberry Pi 5 (8 GB RAM, NVMe):

• INSERT: ~600 ops/sec
• SELECT: ~900 ops/sec