Performance tuning for Open Liberty
Open Liberty supports various parameters and configuration options to influence application performance. You can tune settings for the Java virtual machine (JVM), the transport channel services, and more to improve performance.
The following sections provide more information about tuning parameters and attributes.
Tune the JVM
Tuning the Java virtual machine is the most important tuning step for configuring an environment. To tune the JVM for Open Liberty, you can specify JVM arguments, one option per line, in the jvm.options file in the ${server.config.dir} directory. You can use jvm.options files at the runtime and server levels to specify more server startup options.
The following example specifies one JVM option per line to set maximum heap size to 1024m, set an example system property, and enable verbose garbage collection.
-Xmx1024m
-Dcom.ibm.example.system.property=ExampleValue
-verbose:gcFor faster server startup in a development environment, consider setting the minimum heap size to a small value and the maximum heap size to whatever value is needed for your application. For a production environment, setting the minimum heap size and maximum heap size to the same value can provide the best performance by avoiding heap expansion and contraction.
For more information on the Open Liberty jvm.options file, see jvm.options.
For more information on the JVM tuning options for IBM Java, see Eclipse OpenJ9.
Tune transport channel services
The transport channel services manage client connections, I/O processing for HTTP, thread pools, and connection pools. You can tune the following attributes to improve runtime performance, scalability, or both. In the following table, the examples show you how to code these options in the server.xml file.
| Server configuration | Attribute | Example | Description |
|---|---|---|---|
| <httpOptions maxKeepAliveRequests="-1" /> | This option specifies the maximum number of persistent requests that are allowed on a single HTTP connection if persistent connections are enabled. A value of -1 means unlimited. This option supports low latency or high throughput applications, and TLS connections for use in situations where building up a new connection can be costly. | |
| <connectionManager ... maxPoolSize="40" /> | This option specifies the maximum number of physical connections for the connection pool. The default value is 50. The optimal setting here depends on the application characteristics. For an application in which every thread obtains a connection to the database, you might start with a 1:1 mapping to the | |
| <connectionManager ... purgePolicy="FailingConnectionOnly" /> | This option specifies which connections to end when a stale connection is detected in a pool. The default value is the entire pool. In some cases, you might want to purge only the failing connection by specifying the | |
| <connectionManager ... numConnectionsPerThreadLocal="1" /> | This option specifies the number of database connections to cache for each executor thread. This setting can provide a major improvement on large multi-core machines by reserving the specified number of database connections for each thread. Using thread-local storage for connections can increase performance for applications on multi-threaded systems. When you set | |
| <dataSource ... statementCacheSize="60" > | This option specifies the maximum number of cached prepared statements per connection. To set this option, review the application code (or an SQL trace that you gather from the database or database driver) for all unique prepared statements. Ensure that the cache size is larger than the number of statements. | |
| <dataSource ... isolationLevel="TRANSACTION_READ_COMMITTED"> | The data source isolation level specifies the degree of data integrity and concurrency, which in turn controls the level of database locking. Different options are available for the default transaction isolation level. For more information, see |
Tune the default executor
The Open Liberty default executor is self-tuning and adapts to the current workload by dynamically adding or removing threads. For most workloads, the executor does not require any tuning, and you are advised not to change any settings of the executor unless you encounter specific problems with thread creation. For more information, see Thread pool tuning.
Decrease response time of servlets
To decrease response time of servlets, add the following attribute to the server.xml file.
<webContainer skipMetaInfResourcesProcessing="true"/>This setting prevents the server from searching the meta-inf directory for application resources.
Reduce idle server CPU time
By default, Open Liberty monitors for configuration and application changes by periodically checking for file system updates in the relevant areas of the server installation. For more information on the default settings and functions for configuration and application monitoring, see the following configuration attributes.
To reduce idle server CPU time, add the following attributes to the server.xml file.
<applicationMonitor dropinsEnabled="false" updateTrigger="disabled"/>
<config updateTrigger="disabled"/>When the attributes are added, your server no longer monitors for configuration or application updates.
You can also set the updateTrigger attribute to the MBean value for both the applicationMonitor element and the config element. This setting allows applications and configurations to be updated by an MBean method. However, some amount of CPU time is used. To reduce the amount of CPU time used, you can set the pollingRate attribute of the applicationMonitor element and the monitorInterval attribute of the config element to large values.
The following example shows how to reduce the amount of CPU time that is used when you set the updateTrigger attribute to the value of MBean.
<applicationMonitor updateTrigger="mbean" pollingRate="60s"/>
<config updateTrigger="mbean" monitorInterval="60s"/>Tune startup time
Open Liberty provides convenience features, such as MicroProfile or JakartaEE, which include many individual Liberty features. The convenience features are handy for development environments, when the scope of application function might not be finalized and you want to have all the features readily available. However, each added feature requires server resources during server startup.
You can improve startup times by including only the necessary features in your deployment configuration. A configuration that includes features that are not used by the deployed application might require more CPU, memory, and time to start than necessary. You can use the Open Liberty dev mode to automatically generate a list of the features that your application needs.
By default, the Jakarta Contexts and Dependency Injection (CDI) feature scans all application archives. This feature can increase startup time substantially, particularly in larger applications. You can disable implicit archive scanning by setting the enableImplicitBeanArchives attribute to false. This setting skips the scanning of archives unless they contain a beans.xml file.
<cdi12 enableImplicitBeanArchives="false"/>The CDI feature might be included in your server configuration even if it is not explicitly specified in your server.xml file because other features might implicitly enable it. For example, the MicroProfile feature and the Jakarta EE Web Profile feature each enable the CDI feature by default.
