10 hours ago
- The author argues against the claim that Memcached is better for caching than Redis.
- Redis can disable disk I/O, offering a purely in-memory experience with optional persistence.
- Redis, while single-threaded for data access, can achieve high throughput via pipelining (e.g., 100,000 ops/sec without, 500,000 with pipelining).
- Multiple Redis instances can be used to mimic multi-threaded sharding, similar to Memcached's model.
- Redis's complex operations (e.g., ZRANK, Lua scripts) provide higher 'quality' per operation compared to Memcached's simple GET/SET.
- Redis offers better memory efficiency for small aggregated data types (e.g., integers, hashes) and a more deterministic LRU eviction algorithm than Memcached's slab allocator.
- Redis enables 'smart caching' with data structures like capped lists and sorted sets, going beyond simple key-value storage.
- Redis provides persistence, replication, and observability features (e.g., MONITOR, CLIENT LIST) that are advantageous even for caching.
- Lua scripting in Redis allows server-side processing, reducing data transfer in caching scenarios.
- The author concludes that both tools have merits, but Redis should not be dismissed for caching without considering its advanced features.