Table of Contents
This page contains notes on improving the performance of VuFind in various ways. See the table of contents for a quick overview of topics.
After the initial setup and importing of records, you may notice a significant degradation of performance. This means its time to do some JVM tuning. However, if you look around the Internet, you'll quickly realize performance tuning for Java applications is a bit of a black art. However, we'll provide some basic settings to help improve your performance, as well as resources for more in-depth tuning settings.
As a short side-note, these recommendations are for a “server-class” machine with at least 2 GB RAM and 2 processors. There are limits on the architecture that may require editing the sizes provided in the settings.
This section refers to VuFind 2.x and earlier, which include Solr 4 or earlier. Starting with VuFind 3.0, Solr 5 is bundled with VuFind, and it uses a different startup routine. To adjust the heap memory size in these versions, simply change the SOLR_HEAP environment variable.
If you are using the included distribution of Jetty, the tuning options are set in the JAVA_OPTIONS environmental variable. You can either set this for the user running the instance of Jetty (recommended) or in /etc/profile.
It's recommended that you run the most recent Sun JDK as there is an important switch (-server) that you use that is not in the JRE.
The heap space in Java is the amount of memory allocated to Java for objects. One of the more common issues with a full index of records is an java.lang.OutOfMemoryError exception due to the heap space size. By default, the lower limit is set to 1/64th of the server's physical memory and the max is set at 1/4 the physical memory with a max of 1GB.
It's a good idea to set both of these to the same value as follows (assumes 4GB of RAM):
Note: For very large indices (> 4 million documents) you will need more than 4GB of RAM, and a 64 bit OS to utilize it. You must also tell Java to use the 64 bit “memory model” the setting would then look like:
-d64 -Xms8192m -Xmx8192m
Failure to use the “-d64” flag will result in an error message: “Invalid initial heap size: -Xms8192M The specified size exceeds the maximum representable size. Could not create the Java virtual machine.”
Please note that simply increasing the heap size to a very large one is not recommended, since it may lead to unnecessarily long garbage collection pauses. With very large heaps there's also a caveat: at 32 GB Java switches to a larger pointer size which means that pointers use up more memory than with lower sizes. Therefore it is suggested to keep the heap at below 32 GB or increase it well past it (see this page for more information).
Starting with VuFind 3.0, VuFind includes Solr 5, which does some garbage collection auto-tuning by default; the suggestions below may be less important for these newer releases, and you should probably not tinker with settings unless you have a known problem you are trying to solve.
Garbage collection is how aggressive the JVM is with clearing out unneeded objects. For the purposes of Vufind, parallel garbage collection should work nicely, but read Sun's Tuning Garbage Collection for more in-depth information.
Young generation is a type of garbage collection that has three object spaces, the new object space (Eden) and two survivor spaces. Newer objects are created in Eden, while longer lived objects are moved to the old generation survivor (tenured) spaces. Young generation uses fast copying garbage collection for more frequent clearing of Eden, and more spaced out full garbage collections in the tenured space (which is slower). There are a few switches that will help with this.
Note: if you use the -XX:+UseParNewGC, don't use -XX:+UseParallelGC.
If you want to learn more about your garbage collection performance, you can add the -Xloggc:[filename] parameter to create a log file containing data on Java's behavior. For example, if you want to store logs in your existing Jetty log folder and use the current date as part of the filename, you could do this:
Once you have a log file, you can use the gcviewer or PMAT tool to get a visual representation of its contents. This is very helpful in measuring whether your configuration changes have had a positive impact!
For a good base setting for JAVA_OPTIONS, this is a good set to begin with:
JAVA_OPTIONS="-server -Xmx3800m -Xms3800m -XX:+UseParallelGC -XX:+AggressiveOpts -XX:NewRatio=5"
For more information on Java Tuning, see Java Tuning White Paper and experiment with the settings.
NOTE: The -server switch needs to be the first argument in the string in order for the JDK to pick up the setting.
Be sure to check that the settings are getting applied by running
If all went well, you should see output along the lines of
Checking arguments to VuFind: VUFIND_HOME = /usr/local/vufind SOLR_HOME = /usr/local/vufind/solr SOLR_DATA_DIR = /usr/local/vufind/solr/data JETTY_HOME = /usr/local/vufind/solr/jetty JETTY_LOG = /usr/local/vufind/solr/jetty/logs JETTY_CONF = JETTY_RUN = /tmp JETTY_PID = /tmp/vufind.pid JETTY_CONSOLE = /dev/tty JETTY_PORT = CONFIGS = /usr/local/vufind/solr/jetty/etc/jetty.xml JAVA_OPTIONS = -server -Xmx3800 -Xms3800 -XX:+UseParallelGC -XX:+AggressiveOpts -Dsolr.solr.home=/usr/local/vufind/solr -Dsolr.data.dir=/usr/local/vufind/solr/data -Djetty.logs=/usr/local/vufind/solr/jetty/logs -Djetty.home=/usr/local/vufind/solr/jetty JAVA = /usr/lib/jvm/java-6-sun/bin/java CLASSPATH = RUN_CMD = /usr/lib/jvm/java-6-sun/bin/java -server -Xmx3800 -Xms3800 -XX:+UseParallelGC -XX:+AggressiveOpts -Dsolr.solr.home=/usr/local/vufind/solr -Dsolr.data.dir=/usr/local/vufind/solr/data -Djetty.logs=/usr/local/vufind/solr/jetty/logs -Djetty.home=/usr/local/vufind/solr/jetty -jar /usr/local/vufind/solr/jetty/start.jar /usr/local/vufind/solr/jetty/etc/jetty.xml
User Contributed Settings
William and Mary: Dell Precision 2950, 2×2 dual core Xeon 3.2 GHz processors, 4GB RAM
JAVA_OPTIONS="-server -Xmx3800 -Xms3800 -XX:+UseParallelGC -XX:+AggressiveOpts -XX:NewRatio=5"
Here's a list of useful websites on Java performance tuning:
- Tuning Garbage Collection Outline - older information on the 1.4.2 garbage collection, but still applicable for 1.5+.
- Use Lucene’s MMapDirectory on 64bit platforms, please! - discussion of the MMapDirectory setting, available in Solr 3.x, which may help performance on 64-bit systems
On PHP 5.4, installing a PHP cache like APC can significantly improve performance by reducing the amount of time spent by PHP parsing and compiling code. PHP 5.6 comes with build-In OpCache, which is slightly faster than APC. When upgrading, remember to disable the Apache APC module. OpCache should be configured to 64 MB.
If you created a custom module, remember to create an autoloader_classmap.php by calling
from module directory.
Starting with VuFind 4.1, there is a command-line tool known as the “theme compiler” which can be run to flatten a hierarchy of themes into a single flat theme, reducing the amount of file searching VuFind needs to do in order to find assets and templates.
To use a compiled theme, follow these steps:
1. Run “php $VUFIND_HOME/public/index.php compile theme [your theme] [your compiled theme]” at the command line.
2. Update your config.ini file to use [your compiled theme] instead of [your theme] in the theme setting.
Note: NEVER EDIT THE COMPILED THEME. Instead, edit the source theme and recompile it when you need to make changes. You must use the –force switch of the compiler if you need to overwrite your compiled theme with a new version.
When using database sessions, deleting expired sessions may be a surprisingly heavy process. On a busy site it may be beneficial to turn off PHP's session garbage collection (set session.gc_probability to 0 in php.ini) and run VuFind's expire_sessions utility regularly. This makes sure that garbage collection is done outside Apache that serves user requests and with a method that makes it possible to handle session deletion in a large table.
For a productive environment, you should always enable GZIP compression using mod_deflate. This reduces a amount of data transferred by 70% for text based files like html, js, … To do so,
sudo a2enmod deflate
You should now notice a shrinked transmitted file size in Firefox Developer Tools. You can configure which files are compressed in
Note that the “application/json” type is not usually compressed by default, but turning on GZIP compression for this can significantly approve performance, especially if you use hierarchies and collections.
One thing you should also do is minifying JS and CSS files. This removes whitespaces and line breaks. It can by done by your IDE. Netbeans for example has a plugin Js CSS Minify Compress. The minified files must be configured in theme.config.php and in some cases in the theme files. Just search for “.js” and replace with “.min.js”. If you use LESS to create your CSS files, there is an option –clean-css which optimizes your CSS, too.
Manual minification should not be necessary if you turn on VuFind's built-in asset pipeline (see above).
VuFind's facet feature relies upon Solr's facet support which requires a properly sized Solr “fieldCache”. The solrconfig.xml file configures Solr's fieldCache: see the filtercache section of the Solr documentation for more details. The size of your filterCache should be larger than the number of unique facet values for optimal performance.
Sorting large lists can also use a lot of cache space. You may consider disabling unwanted sort options in order to save memory. Refer to the Solr caching documentation for more details.
Starting with VuFind 2.3, VuFind's default cache sizes have been substantially reduced. The very high settings in earlier versions were known to cause stability and performance problems. It is strongly recommended that you either upgrade or adjust your own Solr configurations. The changes can be viewed in this GitHub commit.
Solr can be configured to perform one or more searches upon startup in order to pre-populate caches and thus allow faster responses. These “warming” searches are configured in solrconfig.xml – look for the <listener> entries for the “firstSearcher” and “newSearcher” events. VuFind's default configuration has some warming searches configured, but you should really customize these for local needs to get better performance. Ideally, the warming searches should exercise all of your facet (and possibly also sort) options. Your base search query can be *:*, or you might want to run a few commonly-requested, known-slow queries. For example, here is Villanova's firstSearcher configuration (and the newSearcher is essentially identical):
<listener event="firstSearcher" class="solr.QuerySenderListener"> <arr name="queries"> <lst> <str name="q">*:*</str> <str name="start">0</str> <str name="rows">10</str> <str name="facet">true</str> <str name="facet.mincount">1</str> <str name="facet.field">collection</str> <str name="facet.field">format</str> <str name="facet.field">publishDate</str> <str name="facet.field">callnumber-first</str> <str name="facet.field">topic_facet</str> <str name="facet.field">authorStr</str> <str name="facet.field">language</str> <str name="facet.field">genre_facet</str> <str name="facet.field">era_facet</str> <str name="facet.field">geographic_facet</str> </lst> </arr> </listener>
Exploiting the System Cache
On Linux, you can sometimes improve performance by loading your entire index into the system cache. This will speed up searches as long as your system has sufficient resources to cache the index.
cat /usr/local/vufind/solr/biblio/index/* > /dev/null
Thanks to Tuan Nguyen for the tip.
Whenever you update the Solr index, it is a good idea to optimize it for better search performance. If you are using spell checking in VuFind 1.0RC2 or later, the optimize operation is also necessary to update your spellchecker index.
Note: Optimizing the index can take a lot of server resources, so you should schedule your index updates and optimizations for non-peak times when possible.
Starting with VuFind 1.0RC2, a simple command-line PHP script is available to optimize the Solr index. Simply switch to the util folder under your VuFind installation and run:
By default the script will optimize the biblio Solr index. For those looking to optimize another index, include the name of the index as an argument:
php optimize.php authority
The optimize action is triggered by posting a simple XML command to the Solr server, so there are many ways to achieve this manually if you do not wish to use the PHP script.
Offloading MARC Records
See the Remote MARC Records page for details on reducing index size by storing MARC records externally to your Solr index. (Note: requires VuFind 2.5 or newer).
Certain operating system limits can impact Solr performance, and starting with release 7.3.1, the software will warn you if you are below the recommended thresholds. See the Linux startup instructions for notes on how to correct this problem.
- Solr Replication - Basic information about making Solr highly available with replication in the context of VuFind.
- Fault Tolerance and Load Balancing - Fault tolerance and load balancing strategies with VuFind.
- Solr Wiki - Performance Factors - Pros and cons of various Solr configuration options; also linked to other helpful wiki pages.
- solrmarc-tech indexing time thread - A discussion which goes into detail on several significant Solr settings.
Search engine crawlers can sometimes put a heavy load on your server, causing performance issues for actual users. The behavior of search engine robots can be controlled with the help of a robots.txt file. See the robots.txt page for more details.
See the Testing Performance page for notes on measuring the performance of your VuFind instance.