AtomicKV: Distributed Key-Value Database

Architected a production-grade, distributed key-value database from scratch. By leveraging a single-threaded asynchronous event loop (epoll), a tiered storage engine (LRU Cache + Custom B-Tree), asynchronous background replication, and a masterless Gossip Protocol architecture, the system achieves massive concurrency, horizontal scalability, and high availability.

Read the full write-up

6
Total Technologies
5
Key Features
Live Demo
Github

Technologies Used

C++
Linux epoll
B-Tree
Gossip Protocol
Bloom Filter
AWS EC2
AtomicKV: Distributed Key-Value Database

Key Features

  • Networking: Engineered a non-blocking network I/O layer utilizing Linux epoll, benchmarking at 10,000+ Requests/Sec on a single thread.
  • Tiered Storage Engine: Designed an L1 RAM LRU Cache protected by a probabilistic Bloom Filter to instantly detect cache misses, falling back to a custom Order-3 B-Tree on disk.
  • Scaling: Implemented a Consistent Hash Ring with virtual nodes for deterministic request routing across the cluster.
  • High Availability: Built a masterless cluster topology using a Gossip Protocol for failure detection and automatic ring rebalancing.
  • Eventual Consistency: Integrated Lamport Clocks for concurrent write conflict resolution (Last Write Wins) and asynchronous background replication.