Redis vs Memcached: Comparative Analysis of Architecture, Memory Management, Persistence, and Clustering

Redis and Memcached are both popular in‑memory data stores, but they differ significantly in design and capabilities. Redis offers richer data structures (list, set, sorted set, hash) and supports master‑slave replication, persistence, and clustering, whereas Memcached focuses on simple key‑value storage with high‑performance O(1) lookups.

1. Network I/O Model – Memcached uses a multithreaded, non‑blocking I/O model built on libevent, which can leverage multiple CPU cores but introduces lock contention for global operations. Redis employs a single‑threaded event‑driven model (aeEvent) using epoll/kqueue/select, providing fast I/O for pure read/write but can become a bottleneck for CPU‑intensive commands.

2. Supported Data Types – Memcached stores only raw key‑value pairs in a large hash table. Redis supports additional structures such as lists, sets, sorted sets, and hashes, enabling more complex use cases.

3. Memory Management – Memcached adopts a slab allocation scheme that pre‑allocates memory chunks of fixed sizes, eliminating fragmentation but causing internal space waste when stored values do not exactly match chunk sizes. Redis uses a custom allocator (zmalloc) that records the size of each allocation, allowing simple free operations and providing statistics via the zmalloc_allocations array; its approach is less complex than slab allocation.

4. Persistence – Memcached is purely in‑memory with no persistence. Redis provides two persistence mechanisms: snapshotting (RDB) and an append‑only file (AOF), enabling data recovery after restarts.

5. Consistency – Memcached offers a CAS (check‑and‑set) command to ensure atomic updates across concurrent clients. Redis does not have CAS but provides transactions (MULTI/EXEC) for atomic execution of command sequences.

6. Cluster Management – Memcached does not include native clustering; distribution is handled by client‑side consistent hashing. Redis includes built‑in clustering (Redis Cluster) with automatic sharding into 4096 hash slots, master‑slave replication for fault tolerance, and optional I/O thread pools to mitigate blocking during swap‑in of evicted values.