Choose Valkey if a BSD-licensed, community-governed Redis-compatible store is a priority. Choose Redis if you need Redis-specific features, its broader ecosystem, or its commercial products. Evaluate Dragonfly if a multithreaded design could let you handle more work per node or reduce shard count. None is a universal winner: test the exact commands, data, and operating setup your application needs.
How to decide between Dragonfly, Redis, and Valkey
Start with the requirement that would be hardest to compromise on, then verify it against the product and version you plan to run. API or protocol compatibility does not prove that every command, module, script, backup format, failover behavior, or operational workflow is interchangeable.
As an Amazon Associate I earn from qualifying purchases.
| Option | Strongest reason to evaluate it | What to verify |
|---|---|---|
| Valkey | BSD licensing, Linux Foundation backing, and familiar Redis-style data structures. | Commands, modules, clients, and operational tooling required by your application. |
| Redis | Redis-specific functionality, its wider ecosystem, or Redis commercial and hosted products. | The exact version’s license terms, required features, service configuration, and total cost. |
| Dragonfly | A multithreaded, shared-nothing design that may suit workloads seeking more throughput per node or fewer shards. | Compatibility and performance with your command mix, data, persistence, failover, and clients. |
For any option, include command mix, value size, concurrency, data-set size, expiration behavior, persistence needs, and tail-latency targets in the evaluation. Also account for replication, clustering, backups, failover, observability, support, and the cost of operating the chosen topology.
When Valkey is the better starting point
Valkey describes itself as a BSD-licensed, in-memory data-structure store for database, cache, message-broker, and streaming-engine workloads. The project identifies Linux Foundation backing. Its documentation lists strings, hashes, lists, sets, sorted sets, bitmaps, HyperLogLogs, geospatial indexes, and streams, along with replication, Lua scripting, eviction, transactions, persistence, Sentinel, and Cluster. It supports persistence as well as cache-only operation.
The Valkey homepage accessed October 7, 2026, lists Valkey 9.1.2, released September 1, 2026. These details describe the project at that time; check the current project documentation when selecting a deployment version.
Valkey is a sensible initial candidate when licensing and community governance matter and the application relies on Redis-style data structures. Treat compatibility as something to test, not an assumption: check each command, module, client behavior, and operational dependency your stack actually uses.
When Redis is the better starting point
Redis remains a current product and ecosystem. Its official About page, accessed October 7, 2026, presents Redis 8.8, Redis Cloud, Redis Software, and workload areas including caching, streaming, session management, search, and feature stores.
Do these 3 things before closing this tab:
1Repair Windows errors before they cause bigger problems2Scan for outdated or missing drivers - takes under a minute3Clear out junk files and repair common Windows errorsRedis can be the more direct fit when an application needs Redis-specific functionality, depends on its ecosystem, or calls for Redis’s commercial support or hosted products. Review the official licensing terms for the exact version and use case before making a licensing decision; do not assume terms are identical across versions or deployments.
When Dragonfly is worth evaluating
Dragonfly describes its datastore as compatible with Redis and Memcached APIs and built on a multithreaded, shared-nothing architecture. That makes it worth testing when a workload may benefit from using more cores in a node, or when fewer Redis-style shards could simplify operations. API compatibility is a reason to evaluate Dragonfly, not evidence that every feature or migration step is identical.
Dragonfly’s documentation claims performance up to 25 times that of Redis. That is a vendor claim, not a general performance guarantee. The vendor’s published comparisons use particular hardware and workload setups, so reproduce relevant tests with your own commands, data, client load, and latency requirements before using the results to size a system.
Rank #3
What Dragonfly’s published benchmarks do—and do not—show
Dragonfly’s comparison page reports YCSB results on AWS c6gn.16xlarge instances with 64 vCPUs and 128 GB of RAM. The figures below are Dragonfly-published results for those stated tests, not independent measurements or expected gains for other deployments.
Free tools Windows power users keep installed
One-click scans. No signup required.
| YCSB workload | Redis result reported by Dragonfly | Dragonfly result reported by Dragonfly |
|---|---|---|
| Write-heavy SET | 125,000 QPS | 3.1 million QPS |
| Read-heavy GET | 240,000 QPS | 4.2 million QPS |
| Mixed 80/20 | 185,000 QPS | 3.7 million QPS |
Results from another Dragonfly-repository benchmark illustrate why hardware and test design matter. Its memtier examples show results near parity for the displayed SET and GET cases on m5.large, then a widening throughput difference on m5.xlarge. The repository’s c6gn.16xlarge discussion reports more than 3.8 million QPS under its setup. These are vendor-documented tests, not a neutral comparison across current versions of all three products.
The same repository describes a roughly 5 GB data-set test using debug populate, update traffic during bgsave, and a claimed 30% improvement in idle-state memory efficiency for Dragonfly. It also reports Redis memory rising to nearly three times Dragonfly’s during snapshotting. Treat those figures as findings from that particular vendor test, not a general memory-saving estimate.
For a useful comparison, reproduce the conditions that affect your system: instance size, client concurrency, threads, pipeline mode, value size, data-set size, command mix, persistence activity, and latency targets. Throughput alone may not answer whether a datastore meets your requirements.
How to check compatibility before migrating
“Redis-compatible” may refer to an API or protocol while leaving differences in commands, modules, file formats, persistence, cluster topology, failover, or operational tools. Valkey documents Redis-like data structures and features; Dragonfly documents Redis and Memcached API compatibility. The documented information does not establish full command-by-command parity among all three.
- Inventory dependencies. Record application commands, Lua scripts, modules, client-library versions, persistence settings, monitoring, and administrative tools.
- Check the target version. Confirm that each required feature behaves as your application expects on the exact target version and deployment type. Pay special attention to modules and scripts.
- Test realistic data and traffic. Use a representative data sample and command mix. Include expiration, persistence, and expected concurrency rather than testing only a simple read or write.
- Exercise operations. Verify import and export, backup and restore, replication, failover, observability, and any cluster or sharding workflow your service depends on.
- Plan and test rollback. Validate a rollback path using the exact source and target versions before moving production data. Do not presume that data files or operational procedures transfer unchanged.
Compare hosted services as deployments, not just engine names
The Valkey project identifies Amazon ElastiCache for Valkey and Google Cloud Memorystore for Valkey, alongside other providers and support organizations. Redis presents Redis Cloud, and Dragonfly documentation presents Dragonfly Cloud. These are options to investigate, not endorsements or guarantees of feature parity.
Compare the actual service configurations available to your application. A managed offering may differ from self-managed software in supported engine versions, regions, failover, persistence, maintenance, observability, support, and price. Check those details for the region and workload you plan to run rather than inferring them from the datastore’s name.
A practical selection path
- Put licensing and governance first? Begin with Valkey, then verify required features and the target provider’s version.
- Need Redis-specific capabilities or its commercial ecosystem? Begin with Redis and check version-specific licensing and deployment terms.
- Looking to scale throughput per node or reduce shard count? Benchmark Dragonfly against your current setup and confirm compatibility and operational needs.
- Unsure which constraint dominates? Build a short evaluation around representative data and traffic, and compare latency, throughput, memory, failure recovery, and total operating cost.
There is no established neutral, apples-to-apples benchmark here that compares current versions of all three systems. Make the choice from the requirements and tests that matter to your deployment, not from a universal ranking.
Quick Recap
Product prices and availability are accurate as of the date/time indicated and are subject to change. Any price and availability information displayed on Amazon at the time of purchase will apply.
