Tips for MySQL Performancein 30mn
AFUP
20 Janvier 2015
Olivier ZemragMySQL Architect
About me
Wonderful 18 Years of IT Technology
More than 10 Years with MySQL family Worked with startups large compagnie such as L’OREAL, ALCATEL,ORANGE, ...
MySQL & open source user since 2000
MySQL Consultant in 2006 at ANASKA
MySQL consultant at SUN in 2008
6 years member of MySQL Oracle Team
Now MySQL Architect
Agenda
IntroductionConfiguration
HardwareDesign
MySQL popularity performancehttp://db-engines.com
Find the 'next' Bottleneck Bench your work load
Test test test Monitor monitor monitor
Think cost / effortChange one thing at a time
Test test testMonitor monitor monitor
Configuration
● Buffer pool
● Log file size
● Other variables
Use large Buffer Pool
● The most important variable with InnoDB
Example : for 32G use 20G forinnodb_buffer_pool_size (or even more)
● Be aware off swaping and other settings !!
Tunne Innodb Log files
For .write heavy workloads use bigInnodb_log_file_size
Be aware of time of restart process
Take caution when you change it
Other variables
● innodb_file_per_table=1
● innodb_stat_on_metadata=OFF
● Check :
– Innodb_flush_at_trx_commit (Durability)
– innodb_flush_method (O_DIRECT)
– skip_name_resolved
● Leave all other variables to default unless youknow what you do or change per session only
hardware
● RAM
● CPU
● Storage
CPU
● Faster is better for query response time
● More cores for more concurrency
MEMORY
● Size of the frequently used data (working set)
● Check response time and throughput
Storage
● RAID10
● SSD
● Fusion IO
● SAN
● Mix :– HDD : log-bin, log-files
– SSD : datadir, innodb_undo_directory, tmpdir
Design
● Optmize, purge, archive data
● Proper Indexation
● Proper usage of temporary tables
– Example of usage of performance_schema to findqueries creating temporary tables (version 5.6)
SELECT * FROMevents_statements_summary_by_digest
WHERE SUM_CREATED_TMP_TABLES > 0
Other Tips & Futures
● Optimize your OS, Optimize your Network
● NoSQL with memcached & InnoDB
● NDB (MySQL Cluster) for high write workloads
● Mix & match InnoDB, memcacehd, NDB Cluster
● MySQL Fabric for sharding
