Valkey vs Dragonfly: The High-Throughput Cache Divide

Dragonfly reaches 3.8 million QPS on a single AWS c6gn.16xlarge node, which equals 25 times the throughput of a single-threaded Redis process. This multi-threaded, shared-nothing architecture partitions the keyspace across threads, where each core owns a slice without mutex contention or lock competition. Because Dragonfly uses io_uring for its asynchronous disk I/O, it avoids the massive memory spikes that typically occur in Redis during the BGSAVE process when the system needs to create a snapshot of the data on disk. In contrast, Valkey 9.0 supports 1 billion operations per second per cluster. Valkey 8.x delivers a 230% performance improvement over Valkey 7.2 and uses a multi-threaded I/O architecture. While Dragonfly provides 15 million QPS for GET operations in pipeline mode, Valkey leads on workload patterns that rely heavily on pipeline-batched requests. Dragonfly 1.18 includes object-compression flags that reduce memory usage by 30% for cache-heavy workloads. Dragonfly provides 80% lower memory usage than Redis for the same workload and reaches 10 million QPS for SET operations. Most users find the vertical scaling of Dragonfly more efficient than managing a sharded cluster.

Valkey acts as a drop-in replacement for Redis 7.2, maintaining full protocol compatibility for existing applications. You can switch by simply updating the endpoint in a configuration file. Dragonfly implements the RESP protocol and most common commands, but it lacks support for some advanced module-specific functions. Redis 8 remains the baseline for functionality, providing native JSON, vector search, and time-series modules in its core. However, Redis 8 uses the AGPLv3 license, which necessitates a legal review for SaaS providers exposing the service to external users. Valkey uses the permissive BSD-3 license. Redis 8 also includes up to 87% faster commands and 2x higher throughput than previous versions. If you are running a production environment, you should check if your specific dependency on Redis Stack modules requires a migration to the specialized modules that Valkey provides for its own users. Does any single engine truly eliminate the need for architectural planning? Valkey 9.0 brings Atomic Slot Migrations, Hash Field Expirations, and Numbered Databases in Cluster Mode to help developers manage complex, large-scale deployments without the need to switch to a different database engine. In 2024, Redis moved to a dual RSALv2 and SSPLv1 license, which prompted the Linux Foundation to launch Valkey.

The decision depends on whether you prioritize vertical scaling or open-source stability. Dragonfly wins for high-throughput, self-hosted workloads where one large node should replace a complex sharded cluster. Use Valkey for standard cache, session storage, or rate-limiting needs, as it provides the most stable, open-source option for cloud-managed environments. If your roadmap requires native vector search or complex JSON documents, stick with Redis 8. Pogocache provides another alternative for low-latency needs, claiming better throughput and lower latency than Valkey, Redis, Dragonfly, and Garnet. Pogocache reaches over 100M ops per second when run embedded. Pogocache was started by Josh Baker and supports Memcache, Valkey/Redis, HTTP, and PostgreSQL wire protocols.