Query Interface

Custom API

Memcached provides get/set and related operations. It also allows users to specify how long a cache item should last.

Storage Model

Custom

Memcached stores a hash table. As of May 2017, it uses MurmurHash 3 as the hash function. Values items are slab-allocated.

Joins

Not Supported

Memcached only supports get/set and related operations on a key-value store. There is no support for SQL queries. Therefore, there is no need to support joins.

Data Model

Key/Value

Memcached servers maintain a key-value map. Keys are up to 250 bytes long and values are up to 1 MB large. There are no relational schemas.

Concurrency Control

Not Supported

Memcached doesn't have the concept of transactions. The get/set operations are atomic, though. For example, multiple sets on the same key can be issued concurrently, but only the last will stay.

System Architecture

Shared-Nothing

In Memcached, a client knows all servers. Servers do not communicate with each other. Each key is hashed to determine which server is used to store the key-value pair. If all clients use the same hash function, then a client can read other clients' cached data.

Logging

Not Supported

There is no logging in Memcached, since it is intended for being a caching system. The cache is not persistant between restarts.

Query Compilation

Not Supported

Memcached only supports get/set and related operations on a key-value store. There is no support for SQL queries. Therefore, there is no need for query compilation.

Stored Procedures

Not Supported

Memcached only supports get/set and related operations on a key-value store. There is no support for SQL queries. Therefore, there is no need to support stored procedures.

Indexes

Hash Table

Memcached stores a key-value hash table on every server. As of May 2017, it uses MurmurHash 3 as the hash function.

Views

Not Supported

Memcached only supports get/set and related operations on a key-value store. There is no support for SQL queries. Therefore, there is no need to support views.

Query Execution

Tuple-at-a-Time Model

Storage Architecture

In-Memory

In Memcached, everything is stored in memory, since it is intended for caching objects stored in a disk-oriented database. By storing everything in memory, it achieved the needed latency and speed. The cache is not persitent between restarts, which is an intended feature.