Location via proxy:   [ UP ]  
[Report a bug]   [Manage cookies]                
Scribd
Upload
693 views

MySQL and SSD: Usage Patterns

MySQL can see significant performance gains from SSD storage, up to 7x faster. Advanced configurations like placing only the most frequently accessed data files or logs on SSD can improve performance even further. For very large datasets or I/O requirements beyond a single SSD's capabilities, multiple instances of MySQL distributed across multiple SSDs can scale up performance in a linear fashion, potentially achieving 20x gains over standard HDD storage. Proper configuration of the MySQL server, storage, and workload distribution are needed to fully leverage SSD performance.

Uploaded by

Oleksiy Kovyrin
Copyright
© Attribution Non-Commercial (BY-NC)
Available Formats
Download as PPTX, PDF, TXT or read online on Scribd
693 views

MySQL and SSD: Usage Patterns

MySQL can see significant performance gains from SSD storage, up to 7x faster. Advanced configurations like placing only the most frequently accessed data files or logs on SSD can improve performance even further. For very large datasets or I/O requirements beyond a single SSD's capabilities, multiple instances of MySQL distributed across multiple SSDs can scale up performance in a linear fashion, potentially achieving 20x gains over standard HDD storage. Proper configuration of the MySQL server, storage, and workload distribution are needed to fully leverage SSD performance.

Uploaded by

Oleksiy Kovyrin
Copyright
© Attribution Non-Commercial (BY-NC)
Available Formats
Download as PPTX, PDF, TXT or read online on Scribd
You are on page 1/ 29

MySQL and SSD: Usage Patterns

Date, time, place: MySQL Conference & Expo 2011


12-Apr-2011

Reporter: Vadim Tkachenko


Co-founder, CTO,
Percona Inc
• You can get up to 7x gain running MySQL on SSD
– Even 20x with some tricks
In this talk
• What is best setup of MySQL to get most benefit from
SSD / Flash
What’s inside
• MySQL and SSD: basics
• MySQL and SSD: advanced schemas
•
•
Types of SSD
• SATA
• PCI-E
• SAN
•
Types of SSD
• SATA
– 200-500MB/sec
– Intel X25-M/E, OCZ, Unigen
• PCI-E
– Over 1GB/sec, 70.000 req/sec, under 1ms response
time
– Virident, FusionIO
• SAN
– Violin memory
–

What is this for MySQL ?
• 1GB/sec – 70,000 req/sec
– A lot, but MySQL can’t use that all
–

MySQL basic usage
– Put all data ( ibdata1, ib_log, tables.ibd) on SSD
• 5-7x difference
–
–

MySQL basic usage
– Boring, some recommendation:
• XFS, better with 4k blocks

• Multiple threads
• Percona Server or InnoDB-plugin or MySQL 5.5
– Still uses about 5,000 req/sec, ~200MB/sec
–
–

Can we do better ?
– Single threaded sequential stuff
• InnoDB transactional logs with fsyncs
• Binary logs
• Doublewrite buffer ( with whole ibdata)
– RAID with BBU good place for them
• Up to 45% improvement
–
–

What is in Percona Server for it ?
– innodb_flush_neighbor_pages= ON | OFF
– innodb_log_block_size = 512 | 4096
– innodb_page_size = 4K | 8K | 16K
• Use carefully
– innodb_doublewrite_file
– Innodb_adaptive_checkpoint=keep_average
– innodb_log_file_size > 4GB
–

What if ?
– Data is too large
• Or SSD too small
–

Naive solution
– Most data on regular disks
– Put “hot” ibd files on SSD
• Symlinks from data directory
–

Naive solution
– It works
– But pain to manage and till first “ALTER TABLE”
– Facebook has patch for it
–

Advanced solution: caching
– Data stored on regular disks
• Caching data on SSD
– Flashcache
• Open source
• Developed and maintained by Facebook, deployed in
production
– DirectCache
• Proprietary solution from FusionIO
–
–

FlashCache
– Shows good results
– Stable work in production deployments
– Not much user friendly
–

FlashCache details
– Write-through and write-back modes
– FIFO and LRU block management
– You need to compile kernel module by yourself
– The same choice for ibdata1/ib_logs location

–

ZFS
– ZFS supports SSD caching mode
– Linux native port available
–

Interesting trick
– For temporary tables on disks
– --tmpdir=/mnt/ssd
–

Still not enough space?
– Several SSD / Flash cards in the server
–

Combining cards
– SATA
• Choice good RAID controller, maybe challenge
– PCI-E
• Software RAID

–
–

What about promised 20x ?
• Single MySQL instance is not able to utilize all IO
provided by card
–

Several instances
• Solution obvious:
– Several instances
–

Experiment
• Dell PowerEdge R815
• 4 physical AMD CPUs / 48 cores
• 144GB of RAM
• Virident tachIOn 200GB card
• 48 user connections
–

Experiment
• Dell PowerEdge R815
• 4 physical AMD CPUs / 48 cores
• 144GB of RAM
• Virident tachIOn 200GB card
• Tpcc-mysql workload
– 48 user connections
–

Results


Results conclusions
• With 120GB memory single instance result worse
then with 26GB
– InnoDB contentions problems again
• Two instances allows to improve 1.5x-6x

Multi-instance
• I do not like it
– Management complexity
– Good scripts solve it
• 2 instances seems reasonable
• SAN-like Flash-arrays
–

The end
• Slides will be online.
http://www.percona.com/about-us/presentations/
• vadim@percona.com
• Your questions ?
• We are hiring!
•

You might also like