Facebook was originally built using MySQL as its primary database management system starting in 2004. However, as Facebook grew to over 2 billion users, storing and processing the enormous amounts of data generated became a challenge for MySQL alone. This led Facebook to develop innovative solutions over the years to scale its infrastructure.
The Early Years: 2004-2008
When Facebook was first created in 2004, the founders chose MySQL as the database system to power the platform. MySQL was selected due to its reputation for being fast, reliable, and free to use. As a start-up, these factors were critical in allowing Facebook to launch and grow quickly in its early stages.
In the early years, Facebook was able to scale MySQL to meet its needs. Major optimizations included:
- Sharding – splitting the data across multiple MySQL instances to distribute the load
- Memory caching with Memcached – caching frequent queries in memory to reduce database hits
- Asynchronous replication – allowing reads and writes across different MySQL servers
With these enhancements, Facebook was able to reach over 100 million users by 2008 on MySQL alone. However, further growth began to expose limits in MySQL’s ability to scale.
The Limitations of MySQL
As Facebook continued growing at a rapid pace, its data storage and access patterns started to push MySQL to its limits in several ways:
- Join performance – Analyzing relationships between data points required inefficient table scans
- Read scaling – MySQL vertical scaling reached limits in supporting heavy read loads
- Shard overhead – More shards increased management overhead and made joins harder
- Cache invalidation – Frequent updates meant higher cache miss rates
It became clear that relying solely on MySQL was not sustainable in the long-term for Facebook. They needed a solution that could scale efficiently to billions of users and provide low latency access to frequently changing hot data.
The Hybrid Data Infrastructure
In response to hitting the scaling limits of MySQL, Facebook engineers designed a novel hybrid data infrastructure. While MySQL continued to store the majority of its data, new systems were introduced to optimize storage and access patterns:
Memcached
Memcached was expanded to cache data across thousands of servers. Frequently accessed data like user profiles and friend relationships was served directly from memory to reduce database load.
NoSQL Databases
New NoSQL databases like HBase and Cassandra were leveraged to store and serve specific types of fast-changing data:
- HBase for storage of the News Feed
- Cassandra for storing inbox messages
The linear scalability of these systems allowed growth without sharding complexity.
Data Warehouses
Hive was implemented on top of Hadoop to build massive data warehouses for analytics. Storing data in centralized repositories optimized for analytics workloads avoided bogging down OLTP databases.
Graph Database
The Graph Search feature launched in 2013 was powered by a graph database called Tao, allowing more efficient analysis of the relationships between objects in Facebook’s social graph.
By blending MySQL with systems designed for specific access patterns, Facebook built a diversified data platform able to scale while minimizing complexity.
MySQL at Facebook Today
The hybrid data infrastructure has proven remarkably successful, allowing Facebook to reach over 2 billion users today. But MySQL remains a critical part of the stack powering Facebook’s core functionality.
MySQL’s strengths continue to make it well suited for important Facebook workloads:
- RAM efficiency for cost effectiveness
- Maturity and stability as a
- Relational model for structured storage
- Strong performance on simple queries and writes
MySQL is primarily used today at Facebook for:
- Storing user account data like logins, settings, profiles
- Managing access control and privacy settings
- Structured content like events, pages, groups
- Serving basic data to apps through API calls
And MySQL has continued evolving to improve scalability. Key innovations include:
MySQL sharding
Facebook has sharding down to a science, allowing linear scaling to hundreds of shards with minimal overhead. Automatic rebalancing and master elections make shards highly available.
Asynchronous Galera Cluster
Galera Cluster delivers multi-master availability through synchronous replication. Facebook developed an asynchronous replication mode that provides high availability without sacrificing write performance.
MySQL proxy layer
A proxy layer efficiently routes queries to appropriate shards. The proxy also allows painless shard migrations and zero downtime upgrades.
These enhancements have kept MySQL highly competitive as one of the pillars of Facebook’s infrastructure.
Conclusion
MySQL has remained a crucial component of Facebook’s data infrastructure despite the adoption of NoSQL databases and other data stores. Its versatility, performance, and ability to scale out make it well suited for Facebook’s core workloads like account and access management. While new data technologies have been introduced to overcome specific limitations, MySQL still powers much of the platform behind the scenes.
Optimizations like sharding, caching, and asynchronous replication have enabled MySQL to continue scaling to meet Facebook’s needs. And MySQL’s rapid innovation, strong community, and advanced features ensure it will likely remain a foundational piece of one of the largest applications in the world for years to come.