Redis vs Memcached
This is the only pairing on the site that is not about licences. Both are permissively usable, both are maintained, and neither is going anywhere. The question is narrower and more useful: are you running Redis because you need what Redis does, or because it was the default the last time somebody added a cache? Memcached has no persistence and no data structures, and for a large share of real deployments that is a feature.
Pick Memcached if your entire Redis usage is get, set and expire on strings, you never needed the data to survive a restart, and you would rather run the simplest thing that caches. Stay on Redis if anything in your code touches a list, a sorted set, a stream, a Lua script or one of the Redis 8 modules, or if a restart losing the whole keyspace would be an incident rather than a warm-up.
Side by side
Redis vs Memcached at a glance
| Redis | Memcached | |
|---|---|---|
| Licence | Tri-licence: choose RSALv2, SSPLv1 or AGPLv3 | 3-Clause BSD |
| OSI-approved? | ✓ yes, via the AGPLv3 option | ✓ yes, permissive with no copyleft obligation |
| Licence changed recently? | Twice: source-available in March 2024, AGPLv3 added for Redis 8 in May 2025 | No |
| Data structures | ✓ lists, sets, sorted sets, hashes, streams | ✗ key and value only |
| Persistence | ✓ | ✗ a restart empties the cache |
| Search, JSON, time series, Bloom | ✓ RediSearch, RedisJSON, RedisTimeSeries, RedisBloom, integral since Redis 8 | ✗ out of scope by design |
| Repository activity (checked Aug 2026) | Actively developed by Redis Ltd. | Push on 10 July 2026 |
| GitHub stars | The category's reference implementation | ~14,200 |
| Vendor cloud pricing | Redis Cloud: free to 30 MB, $5/month Essentials, $200/month Pro minimum | No first-party cloud; not verified for this page |
| Best for | Anything past a plain string cache | A plain string cache you want to stop thinking about |
Sources: Redis licences · Redis pricing · Memcached repository. Compiled August 2026.
Where Memcached wins
It does one thing, so there is nothing to configure wrong
No persistence means no snapshot schedule, no append-only file, no decision about what happens to your data on an unclean shutdown, and no surprise disk usage at three in the morning. The cache is disposable by construction. That is a genuinely smaller operational surface than Redis, and smaller surfaces are how you get fewer incidents.
A licence that has never moved
Memcached is 3-Clause BSD and always has been. Redis went source-available in March 2024 and added AGPLv3 for Redis 8 in May 2025, which is a genuine return to an OSI-approved licence but also the second change in roughly a year. If your reason for reading this page is that you no longer want to track someone else's licensing decisions, that is a real difference and it costs you nothing.
Maintained, not merely old
The repository saw a push on 10 July 2026 and sits at roughly 14,200 stars. A project that finished its job and stopped adding features reads as dormant, but dormant and abandoned are different states. For the contrast, look at KeyDB: more stars in circulation at roughly 12,500, no push since 29 May 2024, and not archived, which is the combination that actually leaves you stranded.
Where Redis still wins
The moment your application needs a leaderboard, a rate limiter, a job queue, pub/sub or a stream, the comparison ends. Those are data structures, and rebuilding one on top of a plain key-value cache means writing a small distributed system by accident. Redis holds them natively, which is why it displaced Memcached in the first place, and the argument has not changed.
Redis 8 also folds RediSearch, RedisJSON, RedisTimeSeries and RedisBloom into the product under the same tri-licence. If you are querying JSON documents or running full-text search against your cache, there is nothing in this comparison for you; Memcached is not attempting that job. And Redis is licensed as open source today under the AGPLv3 option, so "I want an OSI-approved licence" is not by itself a reason to leave.
Migrating Redis to Memcached
The path most teams take:
1. Audit the commands, not the config. Grep the codebase for your client library calls and list every distinct Redis command in use. Anything outside get, set, expire and delete is a blocker to resolve before you plan anything else.
2. Decide what a cold cache costs. A restart empties Memcached completely. Work out how long your service takes to warm up under real traffic and whether the origin database survives that stampede.
3. Move one cache, not all of them. Most systems have several Redis uses under one instance. Peel off the plain string cache first and leave anything stateful where it is; a mixed setup is a perfectly reasonable end state.
4. Keep Redis for the rest. This is not a migration you have to finish. If the leaderboard stays on Redis for ever, that is the correct architecture, not a half-done job.
Common questions
FAQ: Redis vs Memcached
When is Memcached enough to replace Redis?
When your Redis usage is get, set and expire on string values, and nothing in your code depends on a list, a sorted set, a stream or a Lua script. Open your client library usage and count the commands you actually call. If that list fits in one line, Memcached does the job with fewer concepts, no persistence layer to reason about and no licence discussion. If a sorted set is holding a leaderboard or a rate limiter, the answer is no and you should not force it.
Does Memcached store data on disk?
No. Memcached is a pure cache with no persistence, which means a restart empties it and every key has to be recomputed from your source of truth. For a cache in front of a database that is usually acceptable, and treating it as strictly disposable removes a whole class of operational worry. It is disqualifying if you were using Redis as a lightweight datastore, a job queue or session storage that has to survive a restart.
Is Memcached still maintained in 2026?
Yes. The repository had a push on 10 July 2026 and carries roughly 14,200 stars. It is worth stating because Memcached's age gets mistaken for abandonment: a project that solved its problem and stopped growing features looks quiet, but quiet and dead are not the same thing. Compare it with KeyDB, which has more attention at roughly 12,500 stars but no push since 29 May 2024.