Redis vs Memcached

Feature comparison

Redis is an in-memory data structure server. Beyond simple key-value, it provides hashes, lists, sets, sorted sets, streams, bitmaps, HyperLogLog, geospatial indexes, pub/sub channels, and vector embeddings with similarity search through Redis Stack. Persistence is optional through RDB snapshots and append-only files, allowing Redis to function as a primary store for some workloads. Redis Cluster shards data across nodes with automatic failover; Redis Sentinel provides high availability for single-shard deployments. Redis 8.x added Raft-based replication and HNSW vector indexes natively.

Memcached is a multi-threaded distributed memory cache with deliberately minimal scope. It stores opaque key-value pairs with a maximum value size of 1 MB by default, supports TTL expiration and basic compare-and-set semantics, and exposes a simple text or binary protocol. There is no persistence, no replication, and no built-in clustering — sharding is handled client-side through consistent hashing. The simplicity is the point: Memcached is one of the most deployed pieces of caching infrastructure precisely because its behaviour is predictable and operationally minimal.

Per-node performance differs by workload shape. Memcached's multi-threaded slab allocator delivers exceptional throughput for small-value workloads on single nodes — frequently outperforming Redis on raw GET/SET benchmarks on the same hardware. Redis is traditionally single-threaded per shard (with IO threads in recent versions), trading throughput per process for predictable latency and richer semantics. At scale, Redis Cluster's automatic sharding closes the throughput gap; Memcached scales through more client-side shards.

For replication, Redis offers primary-replica replication with automatic failover; Memcached has none natively (some forks and managed offerings add it). For persistence, Redis can survive restarts with RDB or AOF; Memcached loses all data on restart by design. For clients, both have broad language support; Redis clients tend to be more featureful given the wider command set.

For vector search supporting AI workloads, Redis Stack and Redis 8.x provide HNSW indexes and similarity search; Memcached has no equivalent. This single capability has shifted many cache-adjacent workloads toward Redis even where simpler caching would have sufficed.

Pricing comparison

Redis Open Source is licensed under the dual RSALv2/SSPL licence as of recent releases; managed Redis Cloud prices by region, memory, and replication topology, typically ranging $5–$2,500 per month for production tiers. AWS ElastiCache for Redis prices by instance class, typically $20–$8,000 per month per node before reservation discount; Azure Cache for Redis and Google Memorystore for Redis price comparably. Upstash offers serverless Redis priced per request, attractive for variable workloads. Memcached is BSD-licensed open source; AWS ElastiCache for Memcached and Google Memorystore for Memcached price by instance class similar to Redis instances; typical production deployments run $20–$2,000 per month per node.

Five-year cost of ownership for a typical mid-sized caching layer is materially similar between the two — $200K–$800K depending on cloud provider and scale. Redis tends higher when persistence or replication is required; Memcached tends lower when raw cache throughput per node is the dominant cost. The primary buying-side caveat for Redis is licence change in 2024 from BSD to RSALv2/SSPL, which prompted the Valkey fork; enterprises with strict permissive-licence requirements should evaluate Valkey, KeyDB, or Garnet as drop-in alternatives. The Memcached caveat is data loss risk: any deployment treating Memcached as more than a cache will lose data on restart. Pricing as of May 2026.

When to choose Redis

Choose Redis when the workload requires data structures beyond strings, when optional persistence allows survival across restarts, when replication or clustering is needed for availability, when pub/sub or streaming semantics fit the application, when sorted sets enable leaderboards or rate limiting, or when vector similarity search supports retrieval-augmented generation. Redis suits session stores with replication, real-time leaderboards, queue and stream processing, feature stores for machine learning, rate limiters, and AI applications combining cache with embedding search in one system.

When to choose Memcached

Choose Memcached when the workload is pure ephemeral caching with simple key-value access, when value sizes are small and ephemeral, when multi-threaded per-node throughput on raw GET/SET is the dominant performance criterion, when operational simplicity outweighs feature breadth, or when the team prefers strict BSD-licensed open source. Memcached suits page caching, fragment caching, session caching where data loss on restart is acceptable, and high-traffic web stacks where the cache layer is intentionally treated as discardable.

Alternatives to both