An educational relational database system written in C++.
BuzzDB is a modular database system designed for teaching. It introduces core database internals step by step, including storage management, indexing, query execution, transactions, recovery, concurrency control, and query optimization.
The repository is organized as a sequence of standalone C++ versions, so each file can highlight one concept.
BuzzDB only requires a C++17-capable compiler.
For environment setup, see:
Compile any standalone BuzzDB version with:
g++ -std=c++17 -O3 -Wall -Werror -Wextra <module_name>.cppFor example:
g++ -std=c++17 -O3 -Wall -Werror -Wextra 104-buzzdb.cpp
./a.outTo include debugging symbols:
g++ -std=c++17 -O3 -Wall -Werror -Wextra -g <module_name>.cppOn some Apple silicon setups, you may need to include <cassert> if your compiler reports:
error: use of undeclared identifier 'assert'
| Area | What It Covers |
|---|---|
| Data types | Field values for INT, FLOAT, and STRING data. |
| Tuples | Tuple storage, serialization, and row-level representation. |
| Pages | SlottedPage layout for fixed-size page storage. |
| Buffer manager | In-memory page caching with LRU-style replacement. |
| Storage manager | Loading, flushing, and persisting pages on disk. |
| Indexing | Hash indexes, B+tree range access, and related index experiments. |
| Query execution | Operators such as scan, select, projection, aggregation, join, and sort. |
| Transactions | Transaction boundaries, updates, aborts, and recovery examples. |
| Recovery | Shadow paging, WAL, ARIES-style logging, page LSNs, checkpoints, and media recovery. |
| Concurrency control | 2PL, deadlocks, isolation levels, timestamp ordering, OCC, MVCC, and SSI. |
| Query optimization | Statistics, cardinality estimation, join ordering, Cascades-style memo search, LIP, and UDF inlining. |
BuzzDB also includes smaller database-adjacent experiments:
- Patricia trie for compact lookup structures.
- Learned indexes using regression and neural models.
- Columnar storage and compression with SIMD-friendly layouts.
- SIMD operations for high-performance data processing experiments.
- Spatial and vector indexes such as R-tree, ND-tree, and HNSW examples.
NN-buzzdb.cpp: self-contained teaching versions of BuzzDB.z-*.cpp: focused experiments and side projects.job_imdb_data/,imdb*.txt: datasets used by query execution and optimization examples.refs/: reference implementations and papers used while building optimizer versions.booking.txt: toy seat-booking workload for transactions and concurrency control.
Please open an issue or pull request if you find a bug, have a teaching improvement, or want to extend one of the modules.
BuzzDB is licensed under the Apache-2.0 License.