Caching

5 open source tools compared. Sorted by stars — scroll down for our analysis.

Tool	Stars	Velocity	Language	License	Score
Dragonfly Modern Redis/Memcached replacement	30.2k	+50/wk	C++	—	69
Valkey Open source Redis fork maintained by Linux Foundation	25.2k	+101/wk	C	—	69
Memcached High-performance distributed memory caching	14.1k	+11/wk	C	BSD 3-Clause "New" or "Revised" License	79
KeyDB Multi-threaded Redis fork with active replication	12.5k	+6/wk	C++	BSD 3-Clause "New" or "Revised" License	77
Garnet High-performance remote cache-store	11.8k	+13/wk	C#	MIT License	77

Our Analysis

Dragonfly30.2k★

Dragonfly claims to be a modern Redis replacement — same API, same commands, but with a multi-threaded architecture that uses every CPU core instead of Redis's single-threaded model. The pitch: 25x more throughput, 80% less memory, drop-in compatible. If you're running Redis and hitting single-thread bottlenecks, Dragonfly is worth benchmarking. It's API-compatible, so migration is theoretically seamless. Valkey is the community fork of Redis after the license change. KeyDB is another multi-threaded Redis fork. Redis itself is still the default with the largest ecosystem. The catch: The BSL-style license means you can't build a competing caching service. Independent benchmarks tell a different story than Dragonfly's marketing — Redis actually beat Dragonfly by 18-40% in some tests. The multi-threaded advantage only matters at very high throughput. For most indie projects, Redis or Valkey on a single core is more than enough, and the ecosystem support is vastly deeper.

Valkey25.2k★

Valkey is Redis without the Redis drama. After Redis switched to a restrictive license, the Linux Foundation forked it and called it Valkey. Same code, same commands, same performance — but genuinely open source (BSD license). It's Redis 7.2 with a future that isn't controlled by one company. If you're using Redis and worried about license compliance, Valkey is the drop-in replacement. Redis is still Redis — more features in the commercial version, but the license restricts competitive cloud services. Dragonfly is a Redis-compatible rewrite claiming 25x throughput. KeyDB was another fork but lost momentum. Memcached is simpler for pure caching. Best for anyone running Redis who wants license certainty. The migration is literally swapping the binary — your data, configs, and clients all work unchanged. The catch: Valkey forked at Redis 7.2, so newer Redis features won't appear. The ecosystem (Redis Modules, RedisJSON, RedisSearch) may not all work with Valkey. AWS and Google back it, which is reassuring but also means it serves their cloud interests. And if you don't care about licensing, Redis still works fine.

Memcached14.1k★

Memcached is the caching layer that refuses to die — and for good reason. It does one thing: store key-value pairs in memory, blisteringly fast, with multi-threaded performance that still beats Redis on pure read/write throughput for simple caching. If you need a straightforward cache in front of your database and nothing else, Memcached is leaner and uses 30% less memory than Redis for equivalent workloads. But be honest with yourself: Redis does everything Memcached does plus data structures, persistence, pub/sub, and clustering. For new projects, Redis (or its open-source fork Valkey) is the default choice. Dragonfly is the newcomer claiming Redis-compatible performance at lower cost. Memcached still makes sense at massive scale where you're caching simple strings and every byte of RAM matters. Facebook built their caching infrastructure on it for a reason. The catch: no persistence. Server restarts mean cold caches. No data structures beyond strings. And the ecosystem has largely moved to Redis, so finding Memcached-specific tooling and community support gets harder every year.

KeyDB12.5k★

KeyDB is Redis with multi-threading bolted on. A fork that adds per-key locking and MVCC for non-blocking reads, pushing throughput to 1 million ops/sec — 2-5x faster than single-threaded Redis on the same hardware. Because it's a fork, every Redis feature, module, and Lua script works unchanged. Dragonfly is the ground-up rewrite that benchmarks even higher (4M ops/sec) but with less Redis feature compatibility. Valkey is the Linux Foundation's community Redis fork. Garnet is Microsoft's .NET-based alternative with impressive benchmarks. Use KeyDB if you need more Redis throughput without changing your code or learning a new system. Active replication — multiple nodes accepting writes simultaneously — is a genuine differentiator over standard Redis replication. The catch: Snapchat acquired KeyDB, and development has stalled since. If long-term maintenance matters to you, Valkey has stronger community backing and a clearer governance model. KeyDB's BSD-3 license is clean, but an unmaintained fork of a fast-moving project is a ticking clock.

Garnet11.8k★

Microsoft's answer to Redis, and the benchmarks are eye-opening. Garnet delivers 10x higher throughput than Dragonfly on GET commands and massively outscales Redis on multi-threaded workloads. It speaks RESP protocol, so your existing Redis clients work without changes. Redis is the default cache everyone knows. Dragonfly is the multi-threaded Redis alternative. KeyDB is Snapchat's multi-threaded fork. Valkey is the Linux Foundation's post-license-change Redis fork. Built on .NET and Tsavorite (Microsoft Research's concurrent key-value store), Garnet is cross-platform and handles both cache and storage workloads. It supports cluster mode with sharding and replication out of the box. The catch: Garnet is young — launched in 2024, and the production track record is thin compared to Redis's 15+ years. The .NET runtime is unusual for infrastructure software, which may complicate your ops team's tooling. Community adoption is still early, so expect fewer tutorials, fewer integrations, and more undiscovered edge cases. Benchmarks are impressive but synthetic — real-world performance depends on your access patterns.