F1: Google's Next Generation RDBMS

Trends in Architecture and Design
4 April 3:00PM - 3:50PM @ Ballroom G

Experience level: 
Intermediate
Duration: 
50 minutes conference
Many of Google's advertising services have historically been backed by MySQL. Over the last several years, we have migrated several of these systems to F1, a hybrid database that combines easy scaling with the strong consistency guarantees and usability of traditional SQL databases. F1 is built on the Spanner storage system, which provides synchronous wide-area replication and dynamic sharding over a strictly enforced hierarchical schema. Building on this foundation, F1 adds a distributed SQL query engine, flexible indexing, and the ability to join to non-Spanner data sources, among other useful features. On the flip side, Spanner's synchronous replication yields significantly higher write latency than the previous MySQL-based system, but we have largely been able to hide this latency from external users through careful restructuring of applications. This talk will describe how F1 and Spanner work, why and how we migrated some applications off MySQL, and how we structure our schemas and applications to make the most of F1's strengths and weaknesses.


Speakers

Site Reliability Engineer, Google
Biography: 
Brian Biskeborn is a Site Reliability Engineer at Google. He has spent the last 2.5 years working on the F1/Spanner database system and is currently tech lead of the SRE team that runs F1 in production.