PROVIDER Β· The classic distributed memory cache

Memcached

Memcached is the earliest widely used distributed memory cache, whose multi-threaded architecture and minimal key-value model deliver high throughput with no learning curve as the standard LAMP cache layer.

Backend & Database United States 2026.11.15
Backend & Database United States The classic distributed memory cache
4.1pts

Pros

4 items
Multi-threaded architecture uses many cores with high per-instance throughput
Minimal implementation and a simple protocol with mature clients in nearly every language
Automatic LRU eviction with a memory cap keeps usage bounded
BSD licensing allows free deployment and distribution

Cons

4 items
No persistence, so a restart wipes the cache and floods the backend
Simple key-value only without lists, sets, or sorted sets
No replication or HA, relying on client sharding and multi-write
No built-in atomics or scripting, forcing multiple round trips

Product Information

memcached
Category Backend & Database
Company Memcached
Service Region United States
Official Website https://memcached.org/
Recommendation Score 4.1 pts
Published 2026-11-15

Detailed Notes

Overview

Memcached was created in 2003 by Brad Fitzpatrick for LiveJournal under the BSD licence and became the earliest widely used distributed memory cache, known for a minimal key-value model and multi-threaded architecture as the standard cache layer in LAMP stacks.

For Backend & Database work its role is unambiguous, since it only caches: string and simple structure storage, multi-threaded request handling, and automatic LRU eviction give high performance with almost no learning curve, while distribution comes from client-side hashing so servers need not talk to each other.

The limits are equally clear, because there is no persistence, no rich data structures or transactions, no replication so a restart loses everything, and no native atomic command suite, so ranking, pub/sub, persistence, or vector needs point to Redis or Valkey.

Core Strengths

  • Multi-threaded architecture uses many cores with high per-instance throughput
  • Minimal implementation and a simple protocol with mature clients in nearly every language
  • Automatic LRU eviction with a memory cap keeps usage bounded
  • BSD licensing allows free deployment and distribution

Product Ecosystem

Cache model

Key-value storage of serialised objects, commonly caching query results and rendered HTML.

Distribution

Clients map keys by consistent hashing across servers, with cache stampedes to handle on expansion.

Operations

No persistence means a cold start after restart, with hit rate, memory use, and connections the key metrics.

Drawbacks

  • No persistence, so a restart wipes the cache and floods the backend
  • Simple key-value only without lists, sets, or sorted sets
  • No replication or HA, relying on client sharding and multi-write
  • No built-in atomics or scripting, forcing multiple round trips

Use Cases

  • Caching database query results (β˜…β˜…β˜…β˜…β˜…) - Simple key-value uses fewest resources
  • Session sharing (β˜…β˜…β˜…β˜…) - Shared state across machines
  • Page fragment and object caching (β˜…β˜…β˜…β˜…) - Serialised storage
  • Leaderboards and persistence (β˜…) - Use Redis or Valkey

Pricing

Plan Price Scope
Memcached open source $0 BSD licensed, self-managed
Managed cache services Per instance-hour Monitoring and automatic failover
Self-built cluster Server cost Sharding and monitoring on you

The software is free with cost in memory and operations, since instance size determines how much data caches, managed plans bill hourly with small instances in the low tens of dollars monthly, and self-hosting means building sharding and monitoring.

FAQ

  • Memcached or Redis? Memcached for pure caching with simplicity and multi-threaded throughput, Redis or Valkey for data structures, persistence, pub/sub, and atomics.

  • How are cache stampedes handled? Expiring hot keys surges the backend, addressed with mutex locks, staggered expiry times, or logical expiry with background refresh.

  • Can it be a database? No, since no persistence is guaranteed and restarts lose data, so it only accelerates the layer in front of a database.

Visit Memcached Website β†’ https://memcached.org/
Visit Website β†—