DStarDB
Built DStarDB, a C++ multithreaded in-memory database optimized for multi-core performance and clean architecture. Implements Redis-style types (strings, hashes, lists, sets, sorted sets, streams with consumer groups), transactions, TTL/LFU eviction, snapshot persistence, and a shared thread-pool event loop. Benchmarks show lower tail latency and improved throughput vs Redis.
The Challenge
Redis is single-threaded by design, which limits throughput on modern multi-core systems. The challenge was to build a concurrent in-memory database that maintains Redis compatibility while significantly improving multi-threaded performance.
The Solution
Implemented a shared thread-pool event loop architecture with lock-free data structures where possible. Used fine-grained locking for complex operations and developed a clean layered architecture separating networking, command parsing, and storage layers.
Try DStarDB
Real Redis-style command set, running entirely in this tab via a Web Worker. Open the REPL and type — every response is timed in microseconds.
Tech Stack
Interested in working together?