Percona Live 2017 Open Source Database Conference

April 24 - 27, 2017

Santa Clara, California

Scale read workload by sharing data files of InnoDB

Scale read workload by sharing data files of InnoDB

 25 April - 1:20 PM - 2:10 PM @ Ballroom F
Experience level: 
50 minutes conference


The original MySQL provides an ability that if innodb_read_only is turned on, you can setup multiple instances based on same data file. But this approach will prevent all data changes, which is unpractical for most applications. In this session, I'll share how we change InnoDB to make it possible to scale out read only workload by sharing data files, but still provides all types of workload that a normal MySQL allows. More specifically, we'll discuess: * Basic Architecture: physical replication, shared data/log files, etc... * Transaction Management on read only instance: MVCC, Isolation, etc.. * Buffer Pool Management: cache invalidation, data consistency between cache and physical files * Failover between primary server and read only server. Note that this talk is all about internal changes of InnoDB, so having a basic knowleges of redo log, undo log, read view and buffer pool management will be helpful.


zhai weixiang's picture

zhai weixiang

Database Developer, Alibaba


I joined Alibaba in June 2011 and work on MySQL since then. My work includes: Improve the performance of MySQL, develop new features according to requests from our DBA team, and also fix bugs we encountered on production environment. Outside working hours , I like to investigate & test new versions of MySQL and file bugs.

Share this talk