Choosing an ElastiCache Engine: Valkey vs Redis OSS vs Memcached
The first screen of creating an ElastiCache cluster asks for an engine: Valkey, Redis OSS, or Memcached. The conclusion up front: new clusters default to Valkey; existing Redis OSS clusters migrate when the price gap (20–33% lower rates) is worth the move; and Memcached survives only in the narrow niche of pure caches that need neither replication nor persistence. The three overlap heavily in protocol and commands, so the choice is less about features than about pricing and the operating model. Why the Redis family is fast in the first place is covered in Why Redis Is Fast. Prices are for us-east-1.
Side-by-side comparison #
| Aspect | Valkey | Redis OSS | Memcached |
|---|---|---|---|
| Lineage | Open-source fork from Redis 7.2 | Open-source Redis | Independent project |
| Node-based rate | ~20% below Redis OSS | Baseline | Comparable to Redis OSS |
| Serverless rate | ~33% below Redis OSS | Baseline | Serverless supported |
| Serverless minimum | 100MB (~$6/month) | 1GB (~$90/month) | 1GB |
| Replication and failover | Yes | Yes | No |
| Persistence (snapshot, AOF) | Yes | Yes | No |
| Data structures, Pub/Sub, scripting | Redis-compatible plus improvements | Yes | Strings-focused |
| Multithreading | Improved I/O threading | Limited | Multithreaded by design |
How Valkey became the default #
When Redis changed its license in 2024, the open-source side forked at Redis 7.2 and produced Valkey. It is developed under the Linux Foundation with AWS as a major contributor, which makes it effectively a first-class engine in ElastiCache. The protocol and commands are Redis-compatible, so client libraries work unchanged, and later releases added improvements of its own, such as multithreaded I/O.
AWS has also set the direction with pricing. On the same node type, Valkey runs about 20% cheaper than Redis OSS, and about 33% cheaper on serverless. With compatible features at a lower rate, the remaining reasons to pick Redis OSS for a new cluster come down to version pinning or organizational standards. Existing Redis OSS clusters can switch engines to Valkey in place, without downtime, from the console.
The serverless minimum: decisive for small caches #
Node-based clusters follow the RDS grammar: pick an instance, pay by the hour, starting around $11 a month for a small cache.t4g.micro. Serverless bills automatically on storage (GB-hours) and throughput (ECPUs), but with a floor.
Redis OSS serverless bills from a 1GB minimum, about $90 a month even if the cache holds almost nothing. Valkey’s minimum is 100MB, about $6 a month. For a small session cache, that is a 15x difference and the most practical reason Valkey wins any start-small serverless plan.
Node-based versus serverless is a traffic-pattern question. Steady usage favors node-based on rate, further reduced by reserved nodes or Database Savings Plans (about 20% off node-based, about 30% off serverless). As covered in RI vs Savings Plans, these are database-side commitments, separate from compute SPs. Spiky or hard-to-forecast workloads favor serverless for the operations it removes.
Where Memcached still fits #
Memcached has no replication, no failover, no persistence, and no rich data structures. In exchange, it is multithreaded by design, uses all cores of a node, and stays simple, which keeps it valid as a pure cache.
The test is whether the data may vanish. If losing a cache node means refilling from the database and nothing more, and nothing irreplaceable (sessions, rankings) lives there, Memcached suffices. The moment you need failover, lists, sets, Pub/Sub, or snapshots, it is Valkey. In practice one of those requirements tends to arrive eventually, so when in doubt, starting on Valkey avoids the rebuild.
Decision order #
- Start new clusters on Valkey: Redis-compatible at 20–33% lower rates; check only whether a specific reason (version dependency, org standard) forces Redis OSS.
- For small or variable workloads, check the serverless minimums: Valkey’s 100MB (~$6/month) versus Redis OSS’s 1GB (~$90/month) decides small caches.
- For steady traffic, go node-based plus commitments: predictable capacity is cheaper on nodes, with reserved nodes or Database SP stacking further discounts.
- Use Memcached only for loss-tolerant caches: any hint of replication, persistence, or data-structure needs means Valkey.
- For existing Redis OSS, price the switch: same setup, 20% lower rate, so clusters with meaningful monthly bills are worth the no-downtime engine switch. Snapshot first and confirm client compatibility as a baseline.
Summary #
- New ElastiCache clusters default to Valkey: Redis-compatible, about 20% cheaper node-based and 33% cheaper serverless, with AWS driving development.
- The serverless minimum splits at Valkey’s 100MB (~$6/month) versus Redis OSS’s 1GB (~$90/month), which decides start-small caches.
- Steady traffic goes node-based with commitments (reserved nodes, Database SP); variable traffic goes serverless for the operational relief.
- Memcached is for pure, loss-tolerant caches only. When requirements are unclear, starting on Valkey avoids a rebuild.
- Existing Redis OSS clusters can cut rates via the in-place engine switch; review the ones with the largest bills first.