f4 (storage system)

f4 (storage system) is a warm Blob storage system developed by Facebook to efficiently store rarely accessed data, such as replicated copies of photos and videos. It was designed to provide high durability and availability while significantly reducing storage costs compared to traditional replication methods. The system was a key component of Facebook's infrastructure for managing its massive cold data footprint before the company's broader transition to erasure coding and other cloud storage solutions.

Overview

The f4 system was created to address the unsustainable storage costs associated with keeping multiple full replicas of rarely accessed user data, a common practice in large-scale web services like Facebook. It introduced a novel erasure coding scheme optimized for warm storage, moving beyond the simple triple replication used in Facebook's earlier Haystack system. By transitioning a portion of its data center storage from hot to warm tiers, Facebook achieved substantial capital expenditure savings. The development and deployment of f4 represented a significant evolution in data center storage architecture for social media companies during the 2010s.

Architecture

The architecture of f4 is built around a cell-based design, where each cell is a logical grouping of storage servers within a single data center. Data is encoded using a XOR-based erasure code that creates parity blocks, allowing the original data to be reconstructed even if several blocks are lost. This design provides durability comparable to triple replication but with a much lower storage overhead. The system integrates with Facebook's broader TAO graph database and Memcached infrastructure for metadata management and coordination. ZooKeeper is used for distributed coordination and failure detection among the storage nodes.

Implementation

f4 was implemented primarily in C++ for performance-critical components, with Python used for various tools and scripts. It was deployed across multiple Facebook data centers, interfacing with the company's existing web server and load balancing infrastructure. The system employed a custom file system layer optimized for large, immutable blobs and leveraged Linux kernel features for efficient disk I/O. Operational tooling for monitoring and repair was built around Facebook's internal metrics and alerting systems, which were crucial for maintaining the required service-level agreement for data durability.

Features

A primary feature of f4 is its efficient erasure coding scheme, which reduces storage overhead by approximately 50% compared to triple replication for the same level of durability. The system supports fast, parallel reconstruction of lost data blocks to maintain high availability in the event of hardware failures. It is designed for immutable objects, meaning data is written once and never modified, simplifying the consistency model. Furthermore, f4 provides transparent integration with Facebook's photo upload and video streaming pipelines, allowing applications to treat it as a reliable backing store without complex logic.

Use cases

The primary use case for f4 was storing the second and subsequent replicas of user photos and videos uploaded to the Facebook platform, which constitute the vast majority of the company's cold data. It was also used for archival data from other Facebook services like Instagram and Messenger, where strict latency requirements were not necessary. The system served as a cost-effective data warehouse backend for certain analytical workloads involving historical user-generated content. The principles behind f4 influenced later cloud storage designs at Microsoft Azure and Amazon S3 for their infrequent access storage tiers.

Category:Distributed data storage Category:Facebook software Category:Object storage