Re: Question about slow storage and InnoDB compression
The server hosting bacula and the database only has one kind of disk: SATA, maybe I should buy a couple of SSD for mysql. I have read all your mails, and still not sure if I should enable innodb compression. My ibfile is 50 GB, though. Regards Maria Questions: 1) Why are you putting your MySQL data on the same volume as your Bacula backups? Bacula does large sequential I/O and MySQL will do random I/O based on teh structure. What you want to do is: 1) you have 5MB InnoDB Log Files, that's a major bottleneck. I would use at 256MB or 512MB x 2 InnoDB log files. 2) dump and import the database using innodb_file_per_table so that optimization will free up space.. 3) are you running Bacula on the server as well? If so, decrease the buffer pool to 1-2GB.. if not bump it up to to 3GB as you need some memory for bacula and 4, this is the most important one: How big is your MySQL data? Its not that big, I figure in the 80-100GB range. Get yourself a pair of 240GB SSDs, mount it locally for MySQL. S On Tue, Sep 13, 2011 at 21:19, Suresh Kuna sureshkumar...@gmail.com wrote: I would recommend to go for a 15K rpm SSD raid-10 to keep the mysql data and add the Barracuda file format with innodb file per table settings, 3 to 4 GB of innodb buffer pool depending the ratio of myisam v/s innodb in your db. Check the current stats and reduce the tmp and heap table size to a lower value, and reduce the remaining buffer's and cache as well. On Tue, Sep 13, 2011 at 9:06 PM, Maria Arrea maria_ar...@gmx.com wrote: Hello I have upgraded our backup server from mysql 5.0.77 to mysql 5.5.15. We are using bacula as backup software, and all the info from backups is stored in a mysql database. Today I have upgraded from mysql 5.0 to 5.5 using IUS repository RPMS and with mysql_upgrade procedure, no problem so far. This backup systems hold the bacula daemon, the mysql server and the backup of other 100 systems (Solaris/Linux/Windows) Our server has 6 GB of ram, 1 quad Intel Xeon E5520 and 46 TB of raid-6 SATA disks (7200 rpm) connected to a Smart Array P812 controller Red Hat Enterprise Linux 5.7 x64. Our mysql has dozens of millions of lines, and we are using InnoDB as storage engine for bacula internal data. We add hundred of thousands lines /day to our mysql (files are incrementally backed up daily from our 100 servers). So, we have a 7-8 concurrent writes (in different lines, of course) , and theorically we only read from mysql when we restore from backup. Daily we launch a cron job that executes an optimize table in each table of our database to compact the database. It takes almost an hour. We are going to increase the memory of the server from 6 to 12 GB in a couple of weeks, and I will change my.cnf to reflect more memory. My actual my.cnf is attached below: These are my questions: - We have real slow storage (raid 6 SATA), but plenty CPU and ram . Should I enable innodb compression to make this mysql faster? - This system is IOPS-constrained for mysql (fine for backup, though). Should I add a SSD only to hold mysql data? - Any additional setting I should use to tune this mysql server? my.cnf content: [client] port = 3306 socket = /var/lib/mysql/mysql.sock [mysqld] innodb_flush_method=O_DIRECT max_connections = 15 wait_timeout = 86400 port = 3306 socket = /var/lib/mysql/mysql.sock key_buffer = 100M max_allowed_packet = 2M table_cache = 2048 sort_buffer_size = 16M read_buffer_size = 16M read_rnd_buffer_size = 12M myisam_sort_buffer_size = 384M query_cache_type=1 query_cache_size=32M thread_cache_size = 16 query_cache_size = 250M thread_concurrency = 6 tmp_table_size = 1024M max_heap_table = 1024M skip-federated innodb_buffer_pool_size= 2500M innodb_additional_mem_pool_size = 32M [mysqldump] max_allowed_packet = 16M [mysql] no-auto-rehash [isamchk] key_buffer = 1250M sort_buffer_size = 384M read_buffer = 8M write_buffer = 8M [myisamchk] key_buffer = 1250M sort_buffer_size = 384M read_buffer = 8M write_buffer = 8M [mysqlhotcopy] interactive-timeout Regards Maria -- Thanks Suresh Kuna MySQL DBA -- The best compliment you could give Pythian for our service is a referral.
Re: Question about slow storage and InnoDB compression
Am 14.09.2011 09:50, schrieb Maria Arrea: I have read all your mails, and still not sure if I should enable innodb compression if you have enough free cpu-ressources and IO is your problem simply yes because the transfer from/to disk will be not so high as uncompressed signature.asc Description: OpenPGP digital signature
Re: Question about slow storage and InnoDB compression
| 0 | 0 | 0 | | 2011-09-14 11:44:16 | | | latin1_swedish_ci | | row_format=COMPRESSED KEY_BLOCK_SIZE=16 | | +++-++---++-+-+--+---++-+-+-+---+--+-+-+ I am still benchmarking, but I see a 15-20% performance gain after enabling compression using bacula gui (bat). Regards Maria - Original Message - From: Maria Arrea Sent: 09/14/11 09:50 AM To: mysql@lists.mysql.com Subject: Re: Question about slow storage and InnoDB compression The server hosting bacula and the database only has one kind of disk: SATA, maybe I should buy a couple of SSD for mysql. I have read all your mails, and still not sure if I should enable innodb compression. My ibfile is 50 GB, though. Regards Maria Questions: 1) Why are you putting your MySQL data on the same volume as your Bacula backups? Bacula does large sequential I/O and MySQL will do random I/O based on teh structure. What you want to do is: 1) you have 5MB InnoDB Log Files, that's a major bottleneck. I would use at 256MB or 512MB x 2 InnoDB log files. 2) dump and import the database using innodb_file_per_table so that optimization will free up space.. 3) are you running Bacula on the server as well? If so, decrease the buffer pool to 1-2GB.. if not bump it up to to 3GB as you need some memory for bacula and 4, this is the most important one: How big is your MySQL data? Its not that big, I figure in the 80-100GB range. Get yourself a pair of 240GB SSDs, mount it locally for MySQL. S On Tue, Sep 13, 2011 at 21:19, Suresh Kuna sureshkumar...@gmail.com wrote: I would recommend to go for a 15K rpm SSD raid-10 to keep the mysql data and add the Barracuda file format with innodb file per table settings, 3 to 4 GB of innodb buffer pool depending the ratio of myisam v/s innodb in your db. Check the current stats and reduce the tmp and heap table size to a lower value, and reduce the remaining buffer's and cache as well. On Tue, Sep 13, 2011 at 9:06 PM, Maria Arrea maria_ar...@gmx.com wrote: Hello I have upgraded our backup server from mysql 5.0.77 to mysql 5.5.15. We are using bacula as backup software, and all the info from backups is stored in a mysql database. Today I have upgraded from mysql 5.0 to 5.5 using IUS repository RPMS and with mysql_upgrade procedure, no problem so far. This backup systems hold the bacula daemon, the mysql server and the backup of other 100 systems (Solaris/Linux/Windows) Our server has 6 GB of ram, 1 quad Intel Xeon E5520 and 46 TB of raid-6 SATA disks (7200 rpm) connected to a Smart Array P812 controller Red Hat Enterprise Linux 5.7 x64. Our mysql has dozens of millions of lines, and we are using InnoDB as storage engine for bacula internal data. We add hundred of thousands lines /day to our mysql (files are incrementally backed up daily from our 100 servers). So, we have a 7-8 concurrent writes (in different lines, of course) , and theorically we only read from mysql when we restore from backup. Daily we launch a cron job that executes an optimize table in each table of our database to compact the database. It takes almost an hour. We are going to increase the memory of the server from 6 to 12 GB in a couple of weeks, and I will change my.cnf to reflect more memory. My actual my.cnf is attached below:These are my questions:- We have real slow storage (raid 6 SATA), but plenty CPU and ram . Should I enable innodb compression to make this mysql faster? - This system is IOPS-constrained for mysql (fine for backup, though). Should I add a SSD only to hold mysql data? - Any additional setting I should use to tune this mysql server? my.cnf content: [client] port = 3306 socket = /var/lib/mysql/mysql.sock [mysqld] innodb_flush_method=O_DIRECT max_connections = 15 wait_timeout = 86400 port = 3306 socket = /var/lib/mysql/mysql.sock key_buffer = 100M max_allowed_packet = 2M table_cache = 2048 sort_buffer_size = 16M read_buffer_size = 16M read_rnd_buffer_size = 12M myisam_sort_buffer_size = 384M query_cache_type=1 query_cache_size=32M thread_cache_size = 16 query_cache_size = 250M thread_concurrency = 6 tmp_table_size = 1024M max_heap_table = 1024Mskip-federated innodb_buffer_pool_size= 2500M innodb_additional_mem_pool_size = 32M [mysqldump] max_allowed_packet = 16M [mysql] no-auto-rehash [isamchk] key_buffer = 1250M sort_buffer_size = 384M read_buffer = 8M write_buffer = 8M [myisamchk] key_buffer = 1250M sort_buffer_size = 384M read_buffer = 8M write_buffer = 8M [mysqlhotcopy] interactive-timeoutRegards Maria -- Thanks Suresh Kuna MySQL DBA -- The best compliment you could give Pythian for our service is a referral.
Re: Question about slow storage and InnoDB compression
Am 14.09.2011 14:50, schrieb Maria Arrea: I have finally enabled compression: I am still benchmarking, but I see a 15-20% performance gain after enabling compression using bacula gui as expected if disk-io is the only bottenleck the same with NTFS-Compression inside a VMware Machine on modern hardware signature.asc Description: OpenPGP digital signature
Re: Question about slow storage and InnoDB compression
-09-14 11:44:16 | | | latin1_swedish_ci | | row_format=COMPRESSED KEY_BLOCK_SIZE=16 | | | Storage | InnoDB | 10 | Compressed | 1 | 16384 | 16384 | 0 | 0 | 0 | 2 | 2011-09-14 11:44:16 | | | latin1_swedish_ci | | row_format=COMPRESSED KEY_BLOCK_SIZE=16 | | | UnsavedFiles | InnoDB | 10 | Compressed | 0 | 0 | 16384 | 0 | 0 | 0 | 1 | 2011-09-14 11:44:16 | | | latin1_swedish_ci | | row_format=COMPRESSED KEY_BLOCK_SIZE=16 | | | Version | InnoDB | 10 | Compressed | 1 | 16384 | 16384 | 0 | 0 | 0 | | 2011-09-14 11:44:16 | | | latin1_swedish_ci | | row_format=COMPRESSED KEY_BLOCK_SIZE=16 | | +++-++---++-+-+--+---++-+-+-+---+--+-+-+ I am still benchmarking, but I see a 15-20% performance gain after enabling compression using bacula gui (bat). Regards Maria - Original Message - From: Maria Arrea Sent: 09/14/11 09:50 AM To: mysql@lists.mysql.com Subject: Re: Question about slow storage and InnoDB compression The server hosting bacula and the database only has one kind of disk: SATA, maybe I should buy a couple of SSD for mysql. I have read all your mails, and still not sure if I should enable innodb compression. My ibfile is 50 GB, though. Regards Maria Questions: 1) Why are you putting your MySQL data on the same volume as your Bacula backups? Bacula does large sequential I/O and MySQL will do random I/O based on teh structure. What you want to do is: 1) you have 5MB InnoDB Log Files, that's a major bottleneck. I would use at 256MB or 512MB x 2 InnoDB log files. 2) dump and import the database using innodb_file_per_table so that optimization will free up space.. 3) are you running Bacula on the server as well? If so, decrease the buffer pool to 1-2GB.. if not bump it up to to 3GB as you need some memory for bacula and 4, this is the most important one: How big is your MySQL data? Its not that big, I figure in the 80-100GB range. Get yourself a pair of 240GB SSDs, mount it locally for MySQL. S On Tue, Sep 13, 2011 at 21:19, Suresh Kuna sureshkumar...@gmail.com wrote: I would recommend to go for a 15K rpm SSD raid-10 to keep the mysql data and add the Barracuda file format with innodb file per table settings, 3 to 4 GB of innodb buffer pool depending the ratio of myisam v/s innodb in your db. Check the current stats and reduce the tmp and heap table size to a lower value, and reduce the remaining buffer's and cache as well. On Tue, Sep 13, 2011 at 9:06 PM, Maria Arrea maria_ar...@gmx.com wrote: Hello I have upgraded our backup server from mysql 5.0.77 to mysql 5.5.15. We are using bacula as backup software, and all the info from backups is stored in a mysql database. Today I have upgraded from mysql 5.0 to 5.5 using IUS repository RPMS and with mysql_upgrade procedure, no problem so far. This backup systems hold the bacula daemon, the mysql server and the backup of other 100 systems (Solaris/Linux/Windows) Our server has 6 GB of ram, 1 quad Intel Xeon E5520 and 46 TB of raid-6 SATA disks (7200 rpm) connected to a Smart Array P812 controller Red Hat Enterprise Linux 5.7 x64. Our mysql has dozens of millions of lines, and we are using InnoDB as storage engine for bacula internal data. We add hundred of thousands lines /day to our mysql (files are incrementally backed up daily from our 100 servers). So, we have a 7-8 concurrent writes (in different lines, of course) , and theorically we only read from mysql when we restore from backup. Daily we launch a cron job that executes an optimize table in each table of our database to compact the database. It takes almost an hour. We are going to increase the memory of the server from 6 to 12 GB in a couple of weeks, and I will change my.cnf to reflect more memory. My actual my.cnf is attached below:These are my questions:- We have real slow storage (raid 6 SATA), but plenty CPU and ram . Should I enable innodb compression to make this mysql faster? - This system is IOPS-constrained for mysql (fine for backup, though). Should I add a SSD only to hold mysql data? - Any additional setting I should use to tune this mysql server? my.cnf content: [client] port = 3306 socket = /var/lib/mysql/mysql.sock[mysqld] innodb_flush_method=O_DIRECT max_connections = 15 wait_timeout = 86400 port = 3306 socket = /var/lib/mysql/mysql.sock key_buffer = 100M max_allowed_packet = 2M table_cache = 2048 sort_buffer_size = 16M read_buffer_size = 16M read_rnd_buffer_size = 12M myisam_sort_buffer_size = 384M query_cache_type=1 query_cache_size=32M thread_cache_size = 16 query_cache_size = 250M thread_concurrency = 6 tmp_table_size = 1024M max_heap_table = 1024Mskip-federated
Question about slow storage and InnoDB compression
Hello I have upgraded our backup server from mysql 5.0.77 to mysql 5.5.15. We are using bacula as backup software, and all the info from backups is stored in a mysql database. Today I have upgraded from mysql 5.0 to 5.5 using IUS repository RPMS and with mysql_upgrade procedure, no problem so far. This backup systems hold the bacula daemon, the mysql server and the backup of other 100 systems (Solaris/Linux/Windows) Our server has 6 GB of ram, 1 quad Intel Xeon E5520 and 46 TB of raid-6 SATA disks (7200 rpm) connected to a Smart Array P812 controller Red Hat Enterprise Linux 5.7 x64. Our mysql has dozens of millions of lines, and we are using InnoDB as storage engine for bacula internal data. We add hundred of thousands lines /day to our mysql (files are incrementally backed up daily from our 100 servers). So, we have a 7-8 concurrent writes (in different lines, of course) , and theorically we only read from mysql when we restore from backup. Daily we launch a cron job that executes an optimize table in each table of our database to compact the database. It takes almost an hour. We are going to increase the memory of the server from 6 to 12 GB in a couple of weeks, and I will change my.cnf to reflect more memory. My actual my.cnf is attached below: These are my questions: - We have real slow storage (raid 6 SATA), but plenty CPU and ram . Should I enable innodb compression to make this mysql faster? - This system is IOPS-constrained for mysql (fine for backup, though). Should I add a SSD only to hold mysql data? - Any additional setting I should use to tune this mysql server? my.cnf content: [client] port = 3306 socket = /var/lib/mysql/mysql.sock [mysqld] innodb_flush_method=O_DIRECT max_connections = 15 wait_timeout = 86400 port = 3306 socket = /var/lib/mysql/mysql.sock key_buffer = 100M max_allowed_packet = 2M table_cache = 2048 sort_buffer_size = 16M read_buffer_size = 16M read_rnd_buffer_size = 12M myisam_sort_buffer_size = 384M query_cache_type=1 query_cache_size=32M thread_cache_size = 16 query_cache_size = 250M thread_concurrency = 6 tmp_table_size = 1024M max_heap_table = 1024M skip-federated innodb_buffer_pool_size= 2500M innodb_additional_mem_pool_size = 32M [mysqldump] max_allowed_packet = 16M [mysql] no-auto-rehash [isamchk] key_buffer = 1250M sort_buffer_size = 384M read_buffer = 8M write_buffer = 8M [myisamchk] key_buffer = 1250M sort_buffer_size = 384M read_buffer = 8M write_buffer = 8M [mysqlhotcopy] interactive-timeout Regards Maria
Re: Question about slow storage and InnoDB compression
I would recommend to go for a 15K rpm SSD raid-10 to keep the mysql data and add the Barracuda file format with innodb file per table settings, 3 to 4 GB of innodb buffer pool depending the ratio of myisam v/s innodb in your db. Check the current stats and reduce the tmp and heap table size to a lower value, and reduce the remaining buffer's and cache as well. On Tue, Sep 13, 2011 at 9:06 PM, Maria Arrea maria_ar...@gmx.com wrote: Hello I have upgraded our backup server from mysql 5.0.77 to mysql 5.5.15. We are using bacula as backup software, and all the info from backups is stored in a mysql database. Today I have upgraded from mysql 5.0 to 5.5 using IUS repository RPMS and with mysql_upgrade procedure, no problem so far. This backup systems hold the bacula daemon, the mysql server and the backup of other 100 systems (Solaris/Linux/Windows) Our server has 6 GB of ram, 1 quad Intel Xeon E5520 and 46 TB of raid-6 SATA disks (7200 rpm) connected to a Smart Array P812 controller Red Hat Enterprise Linux 5.7 x64. Our mysql has dozens of millions of lines, and we are using InnoDB as storage engine for bacula internal data. We add hundred of thousands lines /day to our mysql (files are incrementally backed up daily from our 100 servers). So, we have a 7-8 concurrent writes (in different lines, of course) , and theorically we only read from mysql when we restore from backup. Daily we launch a cron job that executes an optimize table in each table of our database to compact the database. It takes almost an hour. We are going to increase the memory of the server from 6 to 12 GB in a couple of weeks, and I will change my.cnf to reflect more memory. My actual my.cnf is attached below: These are my questions: - We have real slow storage (raid 6 SATA), but plenty CPU and ram . Should I enable innodb compression to make this mysql faster? - This system is IOPS-constrained for mysql (fine for backup, though). Should I add a SSD only to hold mysql data? - Any additional setting I should use to tune this mysql server? my.cnf content: [client] port = 3306 socket = /var/lib/mysql/mysql.sock [mysqld] innodb_flush_method=O_DIRECT max_connections = 15 wait_timeout = 86400 port = 3306 socket = /var/lib/mysql/mysql.sock key_buffer = 100M max_allowed_packet = 2M table_cache = 2048 sort_buffer_size = 16M read_buffer_size = 16M read_rnd_buffer_size = 12M myisam_sort_buffer_size = 384M query_cache_type=1 query_cache_size=32M thread_cache_size = 16 query_cache_size = 250M thread_concurrency = 6 tmp_table_size = 1024M max_heap_table = 1024M skip-federated innodb_buffer_pool_size= 2500M innodb_additional_mem_pool_size = 32M [mysqldump] max_allowed_packet = 16M [mysql] no-auto-rehash [isamchk] key_buffer = 1250M sort_buffer_size = 384M read_buffer = 8M write_buffer = 8M [myisamchk] key_buffer = 1250M sort_buffer_size = 384M read_buffer = 8M write_buffer = 8M [mysqlhotcopy] interactive-timeout Regards Maria -- Thanks Suresh Kuna MySQL DBA
Re: Question about slow storage and InnoDB compression
Thanks for correcting me in the disk stats Singer, A typo error of SSD instead of SAS 15k rpm. Compression may not increase the memory requirements : To minimize I/O and to reduce the need to uncompress a page, at times the buffer pool contains both the compressed and uncompressed form of a database page. To make room for other required database pages, InnoDB may “evict” from the buffer pool an uncompressed page, while leaving the compressed page in memory. Or, if a page has not been accessed in a while, the compressed form of the page may be written to disk, to free space for other data. Thus, at any given time, the buffer pool may contain both the compressed and uncompressed forms of the page, or only the compressed form of the page, or neither. More details and benefits about the barracuda file format can be found in the below url Which helps to know the pros and cons on file format http://dev.mysql.com/doc/refman/5.6/en/glossary.html#glos_antelope http://dev.mysql.com/doc/innodb/1.1/en/glossary.html#glos_barracuda http://www.mysqlperformanceblog.com/2008/04/23/real-life-use-case-for-barracuda-innodb-file-format/ http://dev.mysql.com/doc/innodb/1.1/en/innodb-other-changes-file-formats.html I would go with the Singer suggestions in What you want to do is part. Thanks Suresh Kuna On Wed, Sep 14, 2011 at 7:21 AM, Singer X.J. Wang w...@singerwang.comwrote: Comments: 1) There is no such thing as 15K RPM SSDs... SSDs are NON ROTATIONAL STORAGE, therefore RPMS make no sense.. 2) Upgrading to Barracuda file format isn't really worth it in this case, you're not going to get any real benefits. In your scenario I doubt InnoDB table compression will help, as it will significantly increase your memory requirements as it to keep uncompressed and compressed copies in RAM. Questions: 1) Why are you putting your MySQL data on the same volume as your Bacula backups? Bacula does large sequential I/O and MySQL will do random I/O based on teh structure. What you want to do is: 1) you have 5MB InnoDB Log Files, that's a major bottleneck. I would use at 256MB or 512MB x 2 InnoDB log files. 2) dump and import the database using innodb_file_per_table so that optimization will free up space.. 3) are you running Bacula on the server as well? If so, decrease the buffer pool to 1-2GB.. if not bump it up to to 3GB as you need some memory for bacula and 4, this is the most important one: How big is your MySQL data? Its not that big, I figure in the 80-100GB range. Get yourself a pair of 240GB SSDs, mount it locally for MySQL. S On Tue, Sep 13, 2011 at 21:19, Suresh Kuna sureshkumar...@gmail.comwrote: I would recommend to go for a 15K rpm SSD raid-10 to keep the mysql data and add the Barracuda file format with innodb file per table settings, 3 to 4 GB of innodb buffer pool depending the ratio of myisam v/s innodb in your db. Check the current stats and reduce the tmp and heap table size to a lower value, and reduce the remaining buffer's and cache as well. On Tue, Sep 13, 2011 at 9:06 PM, Maria Arrea maria_ar...@gmx.com wrote: Hello I have upgraded our backup server from mysql 5.0.77 to mysql 5.5.15. We are using bacula as backup software, and all the info from backups is stored in a mysql database. Today I have upgraded from mysql 5.0 to 5.5 using IUS repository RPMS and with mysql_upgrade procedure, no problem so far. This backup systems hold the bacula daemon, the mysql server and the backup of other 100 systems (Solaris/Linux/Windows) Our server has 6 GB of ram, 1 quad Intel Xeon E5520 and 46 TB of raid-6 SATA disks (7200 rpm) connected to a Smart Array P812 controller Red Hat Enterprise Linux 5.7 x64. Our mysql has dozens of millions of lines, and we are using InnoDB as storage engine for bacula internal data. We add hundred of thousands lines /day to our mysql (files are incrementally backed up daily from our 100 servers). So, we have a 7-8 concurrent writes (in different lines, of course) , and theorically we only read from mysql when we restore from backup. Daily we launch a cron job that executes an optimize table in each table of our database to compact the database. It takes almost an hour. We are going to increase the memory of the server from 6 to 12 GB in a couple of weeks, and I will change my.cnf to reflect more memory. My actual my.cnf is attached below: These are my questions: - We have real slow storage (raid 6 SATA), but plenty CPU and ram . Should I enable innodb compression to make this mysql faster? - This system is IOPS-constrained for mysql (fine for backup, though). Should I add a SSD only to hold mysql data? - Any additional setting I should use to tune this mysql server? my.cnf content: [client] port = 3306 socket = /var/lib/mysql/mysql.sock [mysqld] innodb_flush_method=O_DIRECT max_connections = 15 wait_timeout = 86400 port =
