Memcached

View Current Viewing Revision #3 from 04/28/2018 9:37 a.m.

Memcached (pronunciation: mem-cash-dee) is a distributed in-memory key-value storage system, usually used for caching objects in memory to speed up database applications. Typically, users wrap database queries with Memcached get/set operations to cache recently-used objects.

History

Memcached was originally developed by Brad Fitzpatrick from Danga Interactive for LiveJournal. It was originally written in Perl, but is rewritten in C by Anatoly Vorobey. Now Memcached is used widely in many systems, including YouTube, Twitter, Facebook, and Wikipedia. It is also provided in many cloud platform services, including Google App Engine, Microsoft Azure, and Amazon Web Services.

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.

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.

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.

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.

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.

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.

Query Execution

Tuple-at-a-Time Model

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.

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.