97 Things Every SRE Should Know

Chapter 13. When It Comes to Storage, Think Distributed

Salim Virji

Google

Almost every application, whether on a smartphone or running in a web browser, generates and stores data. As SREs, we often have the responsibility for managing the masses of information that guide and inform decisions for applications as wide-ranging as thermostats to traffic-routing to sharing cat pictures. Distributed storage systems have gained popularity because they provide fault-tolerant and reliable approaches to data management and offer a scalable approach to data storage and retrieval.

Distributed storage is distinct from storage appliances, storage arrays, and storage physically attached to the computer using it; distributed storage systems decouple data producers from the physical media that store this data. By spreading the risk of data storage across different physical media, the system provides speed and reliability, two features that are fundamental to providing a good user experience, whether your user is a human excitedly sharing photographs with family and friends around the world or another computer processing data in a pipeline.

Distributed storage enables concurrent client access: as the system writes the data to multiple locations, there’s no single disk head to block all read operations. Additional copies of the data can be made asynchronously to support increased read demand from ...