Valkey replaces Redis as the standard open source cache

The licensing split

Redis Ltd. changed its licensing model in March 2024 to the SSPL and RSALv2. This change removed the permissive BSD license for all versions from 7.4 onwards. Consequently, the Linux Foundation created Valkey to maintain the open source availability of the software. A group of engineers from the Redis community, including maintainers from Alibaba, Tencent, and Huawei, formed the project within eight days of the announcement. AWS, Google Cloud, Oracle, and Ericsson support this project. Valkey 7.2 began as a direct fork, while version 8.0 marked the first significant release. Since its inception 18 months ago, the project has grown to include over 150 contributors and 40 organizations. Valkey carries a permissive BSD 3-Clause license for teams that must avoid the legal complexities of the AGPL license used by newer Redis versions. You likely already know that licensing changes can disrupt entire infrastructure plans. Valkey remains a single-threaded command execution engine, but it uses multi-threaded I/O to handle network tasks. The Linux Foundation hosts Valkey under the CNCF umbrella.

Performance and feature divergence

Valkey 9.0 uses SIMD acceleration for AVX512 and NEON to increase performance. The project also introduced asynchronous I/O threading in version 8.0 to move connection handling onto dedicated threads. The recent hashtable redesign in Valkey 8.0 reduced overhead by 16 bytes per key-value pair and cut overall memory usage by 9% to 10% for 16-byte keys with 8 or 16-byte values. Valkey is a drop-in replacement for Redis 7.2 because it maintains full command and wire compatibility, and this allows engineering teams to avoid the legal complications associated with the AGPL license used by newer Redis versions. However, the Valkey search module functions as a subset of the Redis search capabilities. It lacks autocomplete, synonyms, spellcheck, and relevance ranking. If your application depends on these specific search functions, Valkey will not satisfy your requirements. Developers can still access structured data through modules like valkey-json and valkey-bloom. Valkey also supports LDAP integration for centralized authentication. It manages complex data types including strings, hashes, lists, sets, sorted sets, streams, bitmaps, hyperloglog, and geospatial indexes. Valkey 9.0 also includes hash field expiration and numbered databases in cluster mode.

The competitive landscape

Dragonfly uses a shared-nothing architecture to partition the keyspace across all available CPU cores. This allows one instance to scale vertically on a single large machine with high throughput. Dragonfly delivers 25x more throughput than legacy in-memory datastores and provides higher cache hit rates with lower tail latency. Valkey and Redis follow a different path by using a single core for command execution. Memcached remains a strong choice for simple key-value workloads where you only need to store strings. Memcached can outperform Redis in raw throughput for simple GET and SET operations on multi-core systems by 15% to 30%. Redis carries about 90 bytes of overhead per key, whereas Memcached carries about 60 bytes of overhead per key. Independent testing in March 2026 showed that caching 1 million keys consumed 140 MB on Memcached versus 180 MB on Redis 8.0. Dragonfly is designed for modern, high-performance application workloads. Memcached is a proven, lightweight option for simple string caching.

Valkey supports remote direct memory access for specialized, high-throughput environments. Which engine handles your specific vector workload best?