Redis vs Memcached: Which In‑Memory Store Wins on Performance and Features?

Network I/O Model

Memcached uses a multithreaded, non‑blocking I/O multiplexing model with a listening thread and worker threads; it relies on libevent and incurs lock contention on global structures such as the

stats

command. Redis employs a single‑threaded event‑loop (AeEvent) that wraps epoll/kqueue/select, achieving maximum throughput for pure I/O but can become a bottleneck when CPU‑intensive commands (e.g., sorting, aggregation) block the event loop.

Supported Data Types

Memcached stores only simple key‑value pairs in a large hash table, offering O(1) lookup. Redis supports additional data structures such as list, set, sorted set (zset), and hash, providing richer functionality for a variety of use cases.

Reference: http://blog.csdn.net/u013256816/article/details/51133134

Memory Management Mechanism

Both systems avoid raw

malloc/free

due to fragmentation and overhead. Memcached adopts a slab allocation scheme: memory is pre‑allocated in chunks of fixed sizes (slabs) to eliminate fragmentation, though this can waste space when stored objects do not exactly match chunk sizes.

When a request arrives, Memcached selects the smallest suitable slab class, finds a free chunk, and stores the item. Expired or evicted items free their chunks for reuse.

Redis implements its own allocator (zmalloc) that records the size of each allocation in a header before the returned pointer. This enables easy calculation of the original pointer for

free

and tracking of total used memory via a static

used_memory

variable.

Redis also maintains an array

zmalloc_allocations

indexed by allocation size, counting how many blocks of each size are allocated.

Data Storage and Persistence

Memcached does not provide persistence; all data resides purely in memory. Redis offers two persistence options: snapshotting (RDB) writes the entire dataset to disk at intervals, and Append‑Only File (AOF) logs every write operation for replay.

Data Consistency

Memcached provides a

cas

command to handle compare‑and‑swap semantics for concurrent updates. Redis lacks

cas

but supports transactions (MULTI/EXEC) to guarantee atomic execution of a sequence of commands.

Cluster Management Differences

Memcached is a pure in‑memory cache without built‑in clustering; distributed deployment relies on client‑side consistent hashing. Redis supports native clustering (Redis Cluster) with sharding across up to 4096 hash slots, master‑slave replication for high availability, and automatic failover.

Source: 朱小厮, blog.csdn.net/u013256816/article/details/51146314