SYSTEM Cited by 1 source
DragonflyDB¶
Overview¶
DragonflyDB is a modern in-memory datastore fully compatible with Redis and Memcached APIs but built on a multi-threaded, shared-nothing architecture with novel algorithms and data structures internally. Source: dragonflydb/dragonfly (GitHub), cited in the benchmark-comparison section of sources/2022-07-11-highscalability-stuff-the-internet-says-on-scalability-for-july-11th-2022.
Benchmark datapoint (2022)¶
From the Redis-vs-KeyDB-vs-Dragonfly-vs-Skytable thread on HN:
| Database | Writes / sec | Reads / sec |
|---|---|---|
| Redis | 112,100 | 99,892 |
| KeyDB | 288,931 | 282,997 |
| Dragonfly | 408,322 | 392,446 |
| Skytable | 619,992 | 676,091 |
At the time of publication, Dragonfly was ~3.6× faster than Redis on the measured workload (single-host, the benchmark's scope limitation applies).
Architectural thesis¶
Redis is single-threaded. Dragonfly takes the opposite bet: multi-threaded shared-nothing — each thread owns its own slice of the keyspace, avoiding cross-thread locking and cache-line bouncing. The architecture targets the modern many- core server (Graviton3, AMD EPYC, Sapphire Rapids) where a single-threaded system leaves most of the chip idle.
Wire-protocol compatibility with Redis + Memcached means existing clients work unchanged; Dragonfly positions itself as a drop-in replacement for workloads where Redis's single- threaded throughput has become the bottleneck.
Context: the 2022 database-palooza¶
The roundup frames Dragonfly alongside a spate of other in-memory-datastore launches in 2022:
- KeyDB — multi-threaded Redis fork (predates Dragonfly).
- Dragonfly — multi-threaded Redis + Memcached replacement (this entry).
- Skytable — multi-threaded datastore with its own protocol.
- CloudFlare D1 — SQLite-based SQL database.
- Anna — key-value store explicitly designed for scalability across many orders of magnitude.
The running theme: Redis's dominance is being challenged by multi-core-aware rewrites in a world where vertical CPU scaling has slowed but per-server core counts keep rising.
Related¶
- systems/redis — the reference point it's compared against.
- companies/highscalability.
- sources/2022-07-11-highscalability-stuff-the-internet-says-on-scalability-for-july-11th-2022.