...
| Code Block | ||||
|---|---|---|---|---|
| ||||
# requires to set the thread_stack to 512K in my.cnf enable_sql_procedures = yes # 8 threads is mostly sufficient for up to 500 users # Other rule of thumb is to have 4 threads per cpu core threads = 8 |
Things you can disable in the default configuration and instead run less frequently with a cronjob.
...
| Code Block | ||||
|---|---|---|---|---|
| ||||
# Should be set as high as you can afford to set it. As it is very important to the overall performance of Kopano. As a rule of thumb it should at least be set to 25% of your ram. cache_cell_size = 2048M # caches all objects and their folder hierarchy cache_object_size = 100kb / user # contains all unique id's of objects cache_indexed_object_size = 512kb / user # Up to 500 users per 1M cache_quota_size = 1M # 60 seconds is more than sufficient cache_quota_lifetime = 1 # Up to 500 users per 1M cache_acl_size = 1M # Up to 500 users per 1M cache_store_size = 1M # 25M per 500 users cache_userdetails_size = 25M |
New proposed defaults
...
cache_indexedobject_size = 32M32+M
Mysql configuration
In general we recommend to use mysql 5.6 or higher for performance reasons, and sugges to have the databases stored on either EXT4 or XFS filesystems.
...
| Code Block | ||||
|---|---|---|---|---|
| ||||
[mysqld] # This can be optionally uncommented. (check if supported on your mysql version) transaction-isolation = READ-COMMITTED # Careful this might also needs binlog_format=ROW if you are using a binlog. #verify# if binlog_format=statement. # Should be the minimum amount of kopano(server.cfg) threads in Kopano server.cfg * 3(default 8) + thread_limit(default 40) + 9. max_connections = 2457 # Best value is 32M, for smaller environments 16M is ok too. max_allowed_packet = 32M # should not be set higher than this, since it has to be allocated before being used tmp_table_size = 64M max_heap_table_size = 64M # max size 2M (regardless of size of environment) sort_buffer_size = 2M join_buffer_size = 2M # Good value is mumber of threads configured + 1 # nice writeup about actual scaling of thread_cache_size http://anothermysqldba.blogspot.de/2013/09/mysql-optimization-tip-threadcachesize.html # #verify# number of threads to be dermineds thread_cache_size = 4 # Only available on mysql 5.5 and higher as above. Good figure the amount is cores/2, depending of usage give read more threads. innodb_write_io_threads = 4 innodb_read_io_threads = 4 # (Kopano tables + max mysql connections) * 10, for example (26 + 24) * 10 # #verify# can be checked for example with "mytop". when you have a low "key efficiency" table_cache has to be increased # table_open_cache was known as table_cache in MySQL 5.1.2 and earlier. table_open_cache = 500 # DO NOT SET IT TOO HIGH! https://www.percona.com/blog/2009/11/16/table_cache-negative-scalability/ # In order to use enable_sql_procedures = yes in Kopano server.cfg 512K is needed at least. thread_stack = 512K # RHEL6 do no use any query cache size !! kills performance query_cache_size = 64M query_cache_limit = 2M # On a shared system (running both Kopano 30% and Mysql 50% and 20% for system Apache etc.) only use 30% of total ram. innodb_buffer_pool_size = 2G # On slow storage 100M is recommended # syntax is file_name:file_size[:autoextend[:max:max_file_size]] innodb_data_file_path = ibdata1:50G:autoextend # On fast like SSD storage or multipath innodb_data_file_path = ibdata1:100G;ibdata2:100G:autoextend # increment ibdata with 1000MB each time (default is 4MB) innodb_autoextend_increment = 1000 innodb_file_per_table = OFF # Be careful 1 second of log (high performance = 2 , normal = 1) innodb_flush_log_at_trx_commit = 2 innodb_log_buffer_size = 8M # The threshold compared to innodb ( 25% of buffer pool size at max 1 GB) innodb_log_file_size = 100M innodb_log_files_in_group = 3 innodb_lock_wait_timeout = 120 # in case of remote storage innodb_flush_method = O_DIRECT # otherwise innodb_flush_method = O_DSYNC |
...
| Code Block | ||||
|---|---|---|---|---|
| ||||
# Make sure it is enabled KeepAlive On |
| Code Block | ||||
|---|---|---|---|---|
| ||||
# Should be higher or equal to the ping time as configured in the z-push/config.php (define('PING_INTERVAL', 30);)
KeepAliveTimeout 45
# This should be higher than the amount of concurrent users you expect.
MaxKeepAliveRequests 500
# Should be sufficiently high for all your users to connect.
MaxClients 500 |
When experiencing a high server load both values could be increased, as the 'push' will take place less often.
...
Checking the current locking statistics
...
| Code Block | ||
|---|---|---|
| ||
84 Last allocated locker ID 0x7fffffff Current maximum unused locker ID 9 Number of lock modes 1500 Maximum number of locks possible 1500 Maximum number of lockers possible 1500 Maximum number of lock objects possible 40 Number of lock object partitions 7 Number of current locks 46 Maximum number of locks at any one time 3 Maximum number of locks in any one bucket 0 Maximum number of locks stolen by for an empty partition 0 Maximum number of locks stolen for any one partition 12 Number of current lockers 23 Maximum number of lockers at any one time 7 Number of current lock objects 39 Maximum number of lock objects at any one time |
Verifying tuning
The following tests can be done to analyze the the tuning results.
They should be performed on adequately warmed up systems compare the output of the the following commands before and after the tuning.
Kopano
kopano-cachestat before and after the tuning.
kopano-stats --top check the rtt and qlen value
Here generally the values should be lower and also the duration of peaks should be shorter.
LDAP
Use the script provided here https://github.com/zarafagroupware/zarafa-tools/blob/master/helpers/ldap/ldap_export.sh (requires ldapsearch to be installed).
This would mimic opening op the GAB 50 as an user does.
Use the wrapper below to have it run a couple of times to get the average of that.
| Code Block | ||||
|---|---|---|---|---|
| ||||
#!/bin/bash
count=0
times=50
while [ $count -lt 50 ]
do
./ldap_export.sh > /dev/null
count=$[$count+1]
done |
Adjust the times variable to a value which makes it easier to compare later.
| Code Block |
|---|
# time ./time.sh |
Mysql
Compare the output ot the mysql tuning script being used before and after.
Use either one MySQLTuner-perl or mysql-tuning-primer scripts.
Always keep in mind there will always be recommendations as tuning is an ongoing process.
Additional Tools / Information
These tools can be used to get some additional tuning recommendations, please only run this tools when the database(is warmed up) has been up and running in production for some time days rather than hours.
https://github.com/major/MySQLTuner-perl
https://launchpad.net/mysql-tuning-primer/
Mysql's temp files on ramdisk using tmpfs, or on a ssd disk partition
It might be useful to run /tmp on tmpfs or instruct mysql to run its temp files somewhere on a tmpfs mountpoint, or an ssd disk partition (YMMV).
Be very careful with this !!
| Code Block | ||
|---|---|---|
| ||
[mysqld] tmpdir = /some/mountpoint |
See the links below for more information.
http://everythingmysql.ning.com/profiles/blogs/using-tmpfs-for-mysqls-tmpdir
More background information
How to calculate a more suitable innodb_log_file_size http://www.percona.com/blog/2008/11/21/how-to-calculate-a-good-innodb-log-file-size/
Information regarding binlog_format=ROW is needed when using transaction-isolation = READ-COMMITTED http://bugs.mysql.com/bug.php?id=66619
Nice writeup about actual scaling of thread_cache_size http://anothermysqldba.blogspot.de/2013/09/mysql-optimization-tip-threadcachesize.html
Kopano administrator manual
kopano core administrator manual
Todo: topic to take a look at use/switch to syslog
Notes on using virtualization
Do's and don'ts
Do
- Use paravirtualized drivers when available for both disk and network interfaces.
- Plan your virtualization needs and size accordingly.
Don't
- Use any of the following features of your virtualization platform as they adversely impact Kopano, MySQL or any program which actively use a lot of memory.
- Ballooning
- Scorecard
- Compression
Page Sharing or Memory De-Duplication
Overcommit Memory, do not give out more memory as the host machine has, as this will cause swapping in the host system
- Do not use virtualization just for the sake of it.