Overview
Explore Facebook's distributed data store for the social graph in this conference talk from USENIX ATC '13. Dive into the TAO system, designed to efficiently handle Facebook's demanding workload with a simple data model and API. Learn about its geographically distributed architecture, capable of processing billions of reads and millions of writes per second. Discover how TAO replaced memcache for many data types, its implementation across thousands of machines, and its ability to provide timely access to petabytes of social graph data. Gain insights into the system's challenges, including bottlenecks, leader cache, timelines, read availability, and database maintenance. Understand the constraints, API model, and failure semantics of TAO, as well as its approach to consistency and deep traversal. Conclude with a Q&A session addressing topics such as the use of MySQL, nodes, datastore workload, and more.
Syllabus
Introduction
What is the Social Graph
When do we render
What is TAO
What are the limitations
Resolving data dependencies
Constraints
API
Model
Architecture
Challenges
bottlenecks
leader cache
timeliness
cache
read availability
database maintenance
summary
questions
why MySQL
nodes
datastore
workload
consistency model
failure semantics
deep traversal
Taught by
USENIX