High-throughput MySQL

Mark Callaghan (Facebook), Ryan Mack (Facebook), Ryan McElroy (Facebook) speak at the 2010 O'Reilly MySQL Conference & Expo.

From the official conference description at http://en.oreilly.com/mysql2010/public/schedule/detail/13223

Commodity servers are available with 8 to 32 CPU cores and 10,000 to 100,000 IOPs. MySQL has some work to do to use this capacity efficiently. The database engineering team at Facebook is one of several groups in the MySQL community working to make MySQL faster on hardware like this.

We will describe bottlenecks encountered at Facebook and provide performance results for the changes. We will describe the SQL that reproduces the problems, symptoms from monitoring that identify the problems and the source code in MySQL that causes the problems.

Work done by the Facebook team includes changes to improve thread scheduling, reduce mutex contention, make mutexes more efficient and identify the sources of disk IO.

Category:

Science & Technology

Tags:

• mysql
• mysqlconf
• facebook
• mark callaghan
• ryan mack
• ryan mcelroy
• bottlenecks
• performance
• scaling
• debugging
• thread scheduling
• threads
• mutex
• I/O