Performance recommendations: Difference between revisions

Revision as of 20:27, 8 October 2013

Moodle can be made to perform very well, at small usage levels or scaling up to many thousands of users. The factors involved in performance are basically the same as for any PHP-based database-driven system. When trying to optimize your server, try to focus on the factor which will make the most difference to the user. For example, if you have relatively more users browsing than accessing the database, look to improve the webserver performance.

Obtain a baseline benchmark

Before attempting any optimization, you should obtain a baseline benchmark of the component of the system you are trying to improve. For Linux try LBS and for Windows use the Performance Monitor. Once you have quantitative data about how your system is performing currently, you'll be able to determine if the change you have made has had any real impact.

The overall aim of adjustments to improve performance is to use RAM (cacheing) and to reduce disk-based activity. It is especially important to try to eliminate swap file usage as much as you can. If your system starts swapping, this is a sign that you need more RAM.

The optimization order preference is usually: primary storage (more RAM), secondary storage (faster hard disks/improved hard disk configuration), processor (more and faster).

Scalability

Moodle's design (with clear separation of application layers) allows for strongly scalable setups. (Please check the list of large Moodle installations.)

Large sites usually separate the web server and database onto separate servers, although for smaller installations this is typically not necessary.

It is possible to load-balance a Moodle installation, for example by using more than one webserver. The separate webservers should query the same database and refer to the same filestore area, but otherwise the separation of the application layers is complete enough to make this kind of clustering feasible. Similarly, the database could be a cluster of servers (e.g. a MySQL cluster), but this is not an easy task and you should seek expert support, e.g. from a Moodle Partner.

Server cluster

Using Moodle forum discussions:

Hardware configuration

Note: The fastest and most effective change that you can make to improve performance is to increase the amount of RAM on your web server - get as much as possible (e.g. 4GB or more). Increasing primary memory will reduce the need for processes to swap to disk and will enable your server to handle more users.

Better performance is gained by obtaining the best processor capability you can, i.e. dual or dual core processors. A modern BIOS should allow you to enable hyperthreading, but check if this makes a difference to the overall performance of the processors by using a CPU benchmarking tool.
If you can afford them, use SCSI hard disks instead of SATA drives. SATA drives will increase your system's CPU utilization, whereas SCSI drives have their own integrated processors and come into their own when you have multiple drives. If you must have SATA drives, check that your motherboard and the drives themselves support NCQ (Native Command Queuing).
Purchase hard disks with a low seek time. This will improve the overall speed of your system, especially when accessing Moodle's reports.
Size your swap file correctly. The general advice is to set it to 4 x physical RAM.
Use a RAID disk system. Although there are many different RAID configurations you can create, the following generally works best:
- install a hardware RAID controller (if you can)
- the operating system and swap drive on one set of disks configured as RAID-1.
- Moodle, Web server and Database server on another set of disks configured as RAID-5.
Use gigabit ethernet for improved latency and throughput. This is especially important when you have your webserver and database server separated out on different hosts.
Check the settings on your network card. You may get an improvement in performance by increasing the use of buffers and transmit/receive descriptors (balance this with processor and memory overheads) and off-loading TCP checksum calculation onto the card instead of the OS.
Read this Case Study on a server stress test with 300 users.
See this accompanying report on network traffic and server loads.
See the Moodle.org configuration
Also see this SFSU presentation at Educause (using VMWare): [1]

Operating System

You can use Linux(recommended), Unix-based, Windows or Mac OS X for the server operating system. *nix operating systems generally require less memory than Mac OS X or Windows servers for doing the same task as the server is configured with just a shell interface. Additionally Linux does not have licensing fees attached, but can have a big learning curve if you're used to another operating system. If you have a large number of processors running SMP, you may also want to consider using a highly tuned OS such as Solaris.
Check your own OS and vendor specific instructions for optimization steps.
- For Linux look at the Linux Performance Team site.
- For Linux investigate the hdparm command, e.g. hdparm -m16 -d1 can be used to enable read/write on multiple sectors and DMA. Mount disks with the async and noatime options.
- For Windows set the sever to be optimized for network applications (Control Panel, Network Connections, LAN connection, Properties, File & Printer Sharing for Microsoft Networks, Properties, Optimization). You can also search the Microsoft TechNet site for optimization documents.

Web server performance

Installing Firefox and the firebug extension will allow you to watch the time it takes for each page component to load. Also, the Yslow extension will evaluate your page against Yahoo's 14 rules, full text Best Practices for Speeding Up Your Web Site, ~~(video)~~ for fast loading websites.

PHP performance

You are strongly recommended to use a PHP accelerator to ease CPU load, such as APC, PHPA, Xcache, WinCache or eAccelerator. (Take care to choose a PHP accelerator that is known to work well with your version of PHP and note that Turck MMCache is no longer maintained and can cause failures with PHP 5).
Improvements in read/write performance can be improved by putting the cached PHP pages on a TMPFS filesystem - but remember that you'll lose the cache contents when there is a power failure or the server is rebooted.
Performance of PHP is better when installed as an Apache/IIS6 ISAPI module (rather than a CGI). IIS 7.0/7.5 (Windows Server 2008/R2) users should choose a FastCGI installation for best performance.
Also check the memory_limit in php.ini, reduce it to 16M for Moodle version earlier than 1.7 (See this forum discussion). For Moodle 1.7 or later, it is recommended that the value of memory_limit should be 40M. As of PHP 5.2.1 the default value for the memory_limit directive is 128M.
Also see PHP_settings_by_Moodle_version

Install HowTo

Apache performance

If you are using Apache on a Windows server, use the build from Apache Lounge which is reported to have performance and stability improvements compared to the official Apache download. Note that this is an unofficial build, so may not keep up with official releases.
Set the MaxClients directive correctly. Use this formula to help (which uses 80% of available memory to leave room for spare):

MaxClients = Total available memory * 80% / Max memory usage of apache process

Memory usage of apache process is usually 10MB but Moodle can easily use up to 100MB per process, so a general rule of thumb is to divide your available memory in megabytes by 100 to get a conservative setting for MaxClients. You are quite likely to find yourself lowering the MaxClients from its default of 150 on a Moodle server. To get a more accurate estimate read the value from the shell command:

#ps -ylC httpd --sort:rss

If you need to increase the value of MaxClients beyond 256, you will also need to set the ServerLimit directive.

Warning: Do not be tempted to set the value of MaxClients higher than your available memory as your server will consume more RAM than available and start to swap to disk.

Consider reducing the number of modules that Apache loads in the httpd.conf file to the minumum necessary to reduce the memory needed.
Use the latest version of Apache - Apache 2 has an improved memory model which reduces memory usage further.
For Unix/Linux systems, consider lowering MaxRequestsPerChild in httpd.conf to as low as 20-30 (if you set it any lower the overhead of forking begins to outweigh the benefits).
For a heavily loaded server, consider setting KeepAlive Off (do this only if your Moodle pages do not contain links to resources or uploaded images) or lowering the KeepAliveTimeout to between 2 and 5. The default is 15 (seconds) - the higher the value the more server processes will be kept waiting for possibly idle connections. A more accurate value for KeepAliveTimeout is obtained by observing how long it takes your users to download a page. After altering any of the KeepAlive variables, monitor your CPU utilization as there may be an additional overhead in initiating more worker processes/threads.
As an alternative to using KeepAlive Off, consider setting-up a Reverse Proxy server infront of the Moodle server to cache HTML files with images. You can then return Apache to using keep-alives on the Moodle server.
If you do not use a .htaccess file, set the AllowOverride variable to AllowOverride None to prevent .htaccess lookups.
Set DirectoryIndex correctly so as to avoid content-negotiation. Here's an example from a production server:

DirectoryIndex index.php index.html index.htm

Unless you are doing development work on the server, set ExtendedStatus Off and disable mod_info as well as mod_status.
Leave HostnameLookups Off (as default) to reduce DNS latency.
Consider reducing the value of TimeOut to between 30 to 60 (seconds).
For the Options directive, avoid Options Multiviews as this performs a directory scan. To reduce disk I/O further use

Options -Indexes FollowSymLinks

Caching (absolutely unsupported, Moodle 1.9 only) Please note that this kind of caching creates major problems during upgrades, you need to remove the caching directives one week before any upgrade. Apache can be told to make pages load a lot faster by specifying that the browser should cache some various page elements such as images and reuse them from local memory rather than ask for them again every time a page is requested. How to do this varies slightly between OSes but there are two basic steps:

Install and enable mod_expires - refer to documentation or man pages
Add this code to the virtual server config file within the <directory> section for the root directory (or within the .htaccess file if AllowOverrides is On):

<IfModule mod_expires.c>
 ExpiresActive On
 ExpiresDefault "access plus 1 seconds"
 ExpiresByType text/html "access plus 1 seconds"
 ExpiresByType image/gif "access plus 1 week"
 ExpiresByType image/jpeg "access plus 1 week"
 ExpiresByType image/png "access plus 1 week"
 ExpiresByType text/css "access plus 1 week"
 ExpiresByType text/javascript "access plus 1 week"
 ExpiresByType application/x-javascript "access plus 1 week"
 ExpiresByType text/xml "access plus 1 seconds"
</IfModule>

The effect is to make everything stay in the cache except HTML and XML, which change dynamically. It's possible to gain a several hundred percent decrease in load times this way. Adjust the cache times according to how often your images etc change.

Compression reduces response times by reducing the size of the HTTP response

Install and enable mod_deflate - refer to documentation or man pages
Add this code to the virtual server config file within the <directory> section for the root directory (or within the .htaccess file if AllowOverrides is On):

<ifModule mod_deflate.c>
  AddOutputFilterByType DEFLATE text/html text/plain text/xml
</ifmodule>

More info: www.metaskills.net

IIS performance

All alter this location in the registry:

HKLM\SYSTEM\CurrentControlSet\Services\Inetinfo\Parameters\

The equivalent to KeepAliveTimeout is ListenBackLog (IIS - registry location is HKLM\ SYSTEM\ CurrentControlSet\ Services\ Inetinfo\ Parameters). Set this to between 2 to 5.
Change the MemCacheSize value to adjust the amount of memory (Mb) that IIS will use for its file cache (50% of available memory by default).
Change the MaxCachedFileSize to adjust the maximum size of a file cached in the file cache in bytes. Default is 262,144 (256K).
Create a new DWORD called ObjectCacheTTL to change the length of time (in milliseconds) that objects in the cache are held in memory. Default is 30,000 milliseconds (30 seconds).

Lighttpd, NginX and Cherokee performance

You can increase server performance by using a light-weight webserver like lighttpd, nginx or cherokee in combination with PHP in FastCGI-mode. Lighttpd was originally created as a proof-of-concept[2] to address the C10k problem and while primarily recommended for memory-limited servers, its design origins and asynchronous-IO model make it a suitable and proven[3] alternative HTTP server for high-load websites and web apps, including Moodle. See the MoodleDocs Lighttpd page for additional information, configuration example and links.

Alternatively, both lighttpd and nginx are capable of performing as a load-balancer and/or reverse-proxy to alleviate load on back-end servers[4], providing benefit without requiring an actual software change on existing servers.

Do note that these are likely to be the least tested server environments of all particularly if you are using advanced features such as web services and/or Moodle Networking. They are probably best considered for heavily used Moodle sites with relatively simple configurations.

Database performance

MySQL performance

The following are MySQL specific settings which can be adjusted for better performance in your my.cnf (my.ini in Windows). The file contains a list of settings and their values. To see the current values use these commands

SHOW STATUS;
SHOW VARIABLES;

Important: You must make backups of your database before attempting to change any MySQL server configuration. After any change to the my.cnf, restart mysqld.

If you are able, the MySQLTuner tool can be run against your MySQL server and will calculate appropriate configuration values for most of the following settings based on your current load, status and variables automatically.

Enable the query cache with

query_cache_type = 1.

For most Moodle installs, set the following:

query_cache_size = 36M 
query_cache_min_res_unit = 2K.

The query cache will improve performance if you are doing few updates on the database.

Set the table cache correctly. For Moodle 1.6 set

table_cache = 256 #(table_open_cache in MySQL > 5.1.2)

(min), and for Moodle 1.7 set

table_cache = 512 #(table_open_cache in MySQL > 5.1.2)

(min). The table cache is used by all threads (connections), so monitor the value of opened_tables to further adjust - if opened_tables > 3 * table_cache(table_open_cache in MySQL > 5.1.2) then increase table_cache upto your OS limit. Note also that the figure for table_cache will also change depending on the number of modules and plugins you have installed. Find the number for your server by executing the mysql statement below. Look at the number returned and set table_cache to this value.

mysql>SELECT COUNT(table_name) FROM information_schema.tables WHERE table_schema='yourmoodledbname';

Set the thread cache correctly. Adjust the value so that your thread cache utilization is as close to 100% as possible by this formula:

thread cache utilization (%) = (threads_created / connections) * 100

The key buffer can improve the access speed to Moodle's SELECT queries. The correct size depends on the size of the index files (.myi) and in Moodle 1.6 or later (without any additional modules and plugins), the recommendation for this value is key_buffer_size = 32M. Ideally you want the database to be reading once from the disk for every 100 requests so monitor that the value is suitable for your install by adjusting the value of key_buffer_size so that the following formulas are true:

key_read / key_read_requests < 0.01
key_write / key_write_requests <= 1.0

Set the maximum number of connections so that your users will not see a "Too many connections" message. Be careful that this may have an impact on the total memory used. MySQL connections usually last for milliseconds, so it is unusual even for a heavily loaded server for this value to be over 200.
Manage high burst activity. If your Moodle install uses a lot of quizzes and you are experiencing performance problems (check by monitoring the value of threads_connected - it should not be rising) consider increasing the value of back_log.
Optimize your tables weekly and after upgrading Moodle. It is good practice to also optimize your tables after performing a large data deletion exercise, e.g. at the end of your semester or academic year. This will ensure that index files are up to date. Backup your database first and then use:

mysql>CHECK TABLE mdl_tablename;
mysql>OPTIMIZE TABLE mdl_tablename;

The common tables in Moodle to check are mdl_course_sections, mdl_forum_posts, mdl_log and mdl_sessions (if using dbsessions). Any errors need to be corrected using REPAIR TABLE (see the MySQL manual and this forum script).

Maintain the key distribution. Every month or so it is a good idea to stop the mysql server and run these myisamchk commands.

#myisamchk -a -S /pathtomysql/data/moodledir/*.MYI

Warning: You must stop the mysql database process (mysqld) before running any myisamchk command. If you do not, you risk data loss.

Reduce the number of temporary tables saved to disk. Check this with the created_tmp_disk_tables value. If this is relatively large (>5%) increase tmp_table_size until you see a reduction. Note that this will have an impact on RAM usage.

PostgreSQL performance

There are some good papers around on tuning PostgreSQL (like this one), and Moodle's case does not seem to be different to the general case.

The first thing to recognise is that if you really need to worry about tuning you should be using a separate machine for the database server. If you are not using a separate machine then the answers to many performance questions are substantially muddied by the memory requirements of the rest of the application.

You should probably enable autovacuum, unless you know what you are doing. Many e-learning sites have predictable periods of low use, so disabling autovacuum and running a specific vacuum at those times can be a good option. Or perhaps leave autovacuum running but do a full vacuum weekly in a quiet period.

Set shared_buffers to something reasonable. For versions up to 8.1 my testing has shown that peak performance is almost always obtained with buffers < 10000, so if you are using such a version, and have more than 512M of RAM just set shared_buffers to 10,000 (8MB).

The buffer management had a big overhaul in 8.2 and "reasonable" is now a much larger number. I have not conducted performance tests with 8.2, but the recommendations from others are generally that you should now scale shared_buffers much more with memory and may continue to reap benefits even up to values like 100,000 (80MB). Consider using 1-2% of system RAM.

PostgreSQL will also assume that the operating system is caching its files, so setting effective_cache_size to a reasonable value is also a good idea. A reasonable value will usually be (total RAM - RAM in use by programs). If you are running Linux and leave the system running for a day or two you can look at 'free' and under the 'cached' column you will see what it currently is. Consider taking that number (which is kB) and dividing it by 10 (i.e. allow 20% for other programs cache needs and then divide by 8 to get pages). If you are not using a dedicated database server you will need to decrease that value to account for usage by other programs.

Some other useful parameters that can have positive effects, and the values I would typically set them to on a machine with 4G RAM, are:

work_mem = 10240

That's 10M of RAM to use instead of on-disk sorting and so forth. That can give a big speed increase, but it is per connection and 200 connections * 10M is 2G, so it can theoretically chew up a lot of RAM.

maintenance_work_mem = 163840

That's 160M of RAM which will be used by (e.g.) VACUUM, index rebuild, cluster and so forth. This should only be used periodically and should be freed when those processes exit, so I believe it is well worth while.

max_fsm_pages = 100000
max_fsm_relations = 5000

These are used to hold the free-space map, and if they are too small you will see performance degradation after the database has been operating for some time. The exact numbers to set can be gleaned from the output of VACUUM VERBOSE, which prints the required FSM pages at the end of it's run. The 5x increase seems to be useful for a Moodle installation, from experience.

wal_buffers = 64

These buffers are used for the write-ahead log, and there have been a number of reports on the PostgreSQL mailing lists of improvement from this level of increase.

This is a little out of date now (version 8.0) but still worth a read: http://www.powerpostgresql.com/Docs

And there is lots of good stuff here as well: http://www.varlena.com/GeneralBits/Tidbits/index.php

Based on Andrew McMillan's post at Tuning PostgreSQL forum thread.

Performance of different Moodle modules

Moodle's activity modules, filters, and other plugins can be activated/deactivated. If necessary, you may wish to deactivate some features (such as chat) if not required - but this isn't necessary. Some notes on the performance of certain modules:

The Chat module is said to be a hog in terms of frequent HTTP requests to the main server. This can be reduced by setting the module to use Streamed updates, or, if you're using a Unix-based webserver, by running the chat in daemon mode. When using the Chat module use the configuration settings to tune for your expected load. Pay particular attention to the chat_old_ping and chat_refresh parameters as these can have greatest impact on server load.
The Quiz module is known to stretch database performance. However, it has been getting better in recent versions, and we don't know of any good, up-to-date performance measurements. (Here is a case study from 2007 with 300 quiz users.)
The Moodle Cron task is triggered by calling the script cron.php. If this is called over HTTP (e.g. using wget or curl) it can take a large amount of memory on large installations. If it is called by directly invoking the php command (e.g. php -f /path/to/moodle/directory/admin/cron.php) efficiency can be much improved.
The Recent activities block is consuming to much resources if you have huge number of records mdl_log. this is being tested to optimize the SQL query.

Moodle Image Optimization

The base images delivered in the original Moodle distribution package provide unoptimized graphics, most of which can benefit from lossless recompression utilizing optipng for PNGs, gifsicle for GIFs and jpegoptim for JPGs. Optimized graphics transfer faster and provide a faster perceived response for clients[5], especially distance learners. The following example will recursively optimize (without any loss of quality) all the graphics and image files included in a base Moodle installation directory on a server with the above commands installed and available.

find /example/directory/moodle-1.9 -iname *.png -exec optipng -o7 {} \;
find /example/directory/moodle-1.9 -iname *.gif -exec gifsicle -O2 -b {} \;
find /example/directory/moodle-1.9 -iname *.jpg -exec jpegoptim -p {} \;

Both optipng and gifsicle are provided in the base repositories of most newer Linux distributions; jpegoptim must be downloaded and installed manually.

@@ Line 138: / Line 138: @@
 ==Database performance==
-Moodle contains a script which will display some key database performance statistics from the [http://phplens.com/lens/adodb/docs-perf.htm ADOdb performance monitor]. Run the script in your browser as in the following example:
- http://www.mymoodle.com/admin/dbperformance.php
-Use the data displayed as a guide to tune and improve the performance of your database server.
 ===MySQL performance===

Documentation