|Oracle GlassFish Server Performance Tuning Guide
Part Number E24936-01
The following topics are addressed here:
Java SE 6.0 provides two implementations of the HotSpot Java virtual machine (JVM):
The client VM is tuned for reducing startup time and memory footprint. Invoke it by using the
-client JVM command-line option.
The server VM is designed for maximum program execution speed. Invoke it by using the
-server JVM command-line option.
By default, the GlassFish Server uses the JVM setting appropriate to the purpose:
Developer Profile, targeted at application developers, uses the
-client JVM flag to optimize startup performance and conserve memory resources.
Enterprise Profile, targeted at production deployments, uses the
-server JVM flag to maximize program execution speed.
You can override the default JVM options by following the instructions in "Administering JVM Options" in Oracle GlassFish Server Administration Guide. If using the Administration Console, navigate to the Configurations>configuration-name>JVM Settings node, and then click the JVM Options tab. Refer to the online help for complete information about the settings on this page.
For more information on server-class machine detection in Java SE 6.0, see Server-Class Machine Detection.
For more information on JVMs, see Java Virtual Machines.
In some situations, performance can be improved by using large page sizes. For Ultrasparc CMT systems, include the
-XX:LargePageSizeInBytes=256m arguments with your JVM tuning.
s2 need to be replicated to an instance (backup server), the replication module batches the replication messages to be sent to that instance instead of sending separate replication messages. This improves performance. In configurations in which a lot of session replication is performed, you may find better performance by tuning the
org.shoal.cache.transmitter.max.batch.size system property. This property determines the number of replication messages that constitute one batch.
The default value for this property is
20. You can try setting it as high as
90, depending on system loads. Like all system properties, this property is set with the
-D flag in your Java arguments.
The efficiency of any application depends on how well memory and garbage collection are managed. The following sections provide information on optimizing memory and allocation functions:
Garbage collection (GC) reclaims the heap space previously allocated to objects no longer needed. The process of locating and removing the dead objects can stall any application and consume as much as 25 percent throughput.
Almost all Java Runtime Environments come with a generational object memory system and sophisticated GC algorithms. A generational memory system divides the heap into a few carefully sized partitions called generations. The efficiency of a generational memory system is based on the observation that most of the objects are short lived. As these objects accumulate, a low memory condition occurs forcing GC to take place.
The heap space is divided into the old and the new generation. The new generation includes the new object space (eden), and two survivor spaces. The JVM allocates new objects in the eden space, and moves longer lived objects from the new generation to the old generation.
The young generation uses a fast copying garbage collector which employs two semi-spaces (survivor spaces) in the eden, copying surviving objects from one survivor space to the second. Objects that survive multiple young space collections are tenured, meaning they are copied to the tenured generation. The tenured generation is larger and fills up less quickly. So, it is garbage collected less frequently; and each collection takes longer than a young space only collection. Collecting the tenured space is also referred to as doing a full generation collection.
The frequent young space collections are quick (a few milliseconds), while the full generation collection takes a longer (tens of milliseconds to a few seconds, depending upon the heap size).
Other GC algorithms, such as the Concurrent Mark Sweep (CMS) algorithm, are incremental. They divide the full GC into several incremental pieces. This provides a high probability of small pauses. This process comes with an overhead and is not required for enterprise web applications.
When the new generation fills up, it triggers a minor collection in which the surviving objects are moved to the old generation. When the old generation fills up, it triggers a major collection which involves the entire object heap.
Both HotSpot and Solaris JDK use thread local object allocation pools for lock-free, fast, and scalable object allocation. So, custom object pooling is not often required. Consider pooling only if object construction cost is high and significantly affects execution profiles.
The default collector for Java server class machines will optimize for throughput but be tolerant of somewhat long pause times. If you would prefer to have minimal pause times at the expense of some throughput and increased CPU usage, consider using the CMS collector.
Follow this procedure.
Make sure that the system is not using 100 percent of its CPU.
Configure the CMS collector in the server instance.
To do this, add the following JVM options:
jvmstat utility to monitor HotSpot garbage collection. (See Further Information.)
For detailed information on tuning the garbage collector, see Java SE 6 HotSpot Virtual Machine Garbage Collection Tuning.
The two primary measures of garbage collection performance are throughput and pauses. Throughput is the percentage of the total time spent on other activities apart from GC. Pauses are times when an application appears unresponsive due to GC.
Two other considerations are footprint and promptness. Footprint is the working size of the JVM process, measured in pages and cache lines. Promptness is the time between when an object becomes dead, and when the memory becomes available. This is an important consideration for distributed systems.
A particular generation size makes a trade-off between these four metrics. For example, a large young generation likely maximizes throughput, but at the cost of footprint and promptness. Conversely, using a small young generation and incremental GC will minimize pauses, and thus increase promptness, but decrease throughput.
JVM diagnostic output will display information on pauses due to garbage collection. If you start the server in verbose mode (use the command
asadmin start-domain --verbose domain), then the command line argument
-verbose:gc prints information for every collection. Here is an example of output of the information generated with this JVM flag:
[GC 50650K->21808K(76868K), 0.0478645 secs] [GC 51197K->22305K(76868K), 0.0478645 secs] [GC 52293K->23867K(76868K), 0.0478645 secs] [Full GC 52970K->1690K(76868K), 0.54789968 secs]
On each line, the first number is the combined size of live objects before GC, the second number is the size of live objects after GC, the number in parenthesis is the total available space, which is the total heap minus one of the survivor spaces. The final figure is the amount of time that the GC took. This example shows three minor collections and one major collection. In the first GC, 50650 KB of objects existed before collection and 21808 KB of objects after collection. This means that 28842 KB of objects were dead and collected. The total heap size is 76868 KB. The collection process required 0.0478645 seconds.
Other useful monitoring options include:
-XX:+PrintGCDetails for more detailed logging information
-Xloggc:file to save the information in a log file
To specify the attributes for the Java virtual machine, use the Administration Console and set the property under config-name > JVM settings (JVM options).
For applications that do not dynamically generate and load classes, the size of the permanent generation does not affect GC performance. For applications that dynamically generate and load classes (for example, JSP applications), the size of the permanent generation does affect GC performance, since filling the permanent generation can trigger a Full GC. Tune the maximum permanent generation with the
Although applications can explicitly invoke GC with the
System.gc() method, doing so is a bad idea since this forces major collections, and inhibits scalability on large systems. It is best to disable explicit GC by using the flag
On Windows systems, setting the
-XX:+DisableExplicitGC option might prevent the renaming or removal of open application files. As a result, deployment, redeployment, or other operations that attempt to rename or delete files might fail.
Application files can remain open because the files have been used by class loaders to find classes or resources, or have been opened explicitly by GlassFish Server or application code but never explicitly closed. On Windows systems, open files cannot be renamed or deleted. To overcome this limitation, GlassFish Server uses the
System.gc() method to garbage collect the Java object that corresponds to an open file. When the Java object that corresponds to an open file is garbage collected, the object's
finalize method closes the open channel to the file. GlassFish Server can then delete or rename the file.
GlassFish Server uses RMI in the Administration module for monitoring. Garbage cannot be collected in RMI-based distributed applications without occasional local collections, so RMI forces a periodic full collection. Control the frequency of these collections with the property
-sun.rmi.dgc.client.gcInterval. For example,
- java -Dsun.rmi.dgc.client.gcInterval=3600000 specifies explicit collection once per hour instead of the default rate of once per minute.
This section discusses topics related to tuning the Java Heap for performance.
Maximum heap size depends on maximum address space per process. The following table shows the maximum per-process address values for various platforms:
|Operating System||Maximum Address Space Per Process|
Oracle/Redhat/Ubuntu Linux 32-bit
Oracle/Redhat/Ubuntu Linux 64-bit
Solaris x86 (32-bit)
Maximum heap space is always smaller than maximum address space per process, because the process also needs space for stack, libraries, and so on. To determine the maximum heap space that can be allocated, use a profiling tool to examine the way memory is used. Gauge the maximum stack space the process uses and the amount of memory taken up libraries and other memory structures. The difference between the maximum address space and the total of those values is the amount of memory that can be allocated to the heap.
You can improve performance by increasing your heap size or using a different garbage collector. In general, for long-running server applications, use the Java SE throughput collector on machines with multiple processors (
-XX:+AggressiveHeap) and as large a heap as you can fit in the free memory of your machine.
You can control the heap size with the following JVM parameters:
-Xmx parameters define the minimum and maximum heap sizes, respectively. Since GC occurs when the generations fill up, throughput is inversely proportional to the amount of the memory available. By default, the JVM grows or shrinks the heap at each GC to try to keep the proportion of free space to the living objects at each collection within a specific range. This range is set as a percentage by the parameters
-XX:MaxHeapFreeRatio=maximum; and the total size bounded by
Set the values of
-Xmx equal to each other for a fixed heap size. When the heap grows or shrinks, the JVM must recalculate the old and new generation sizes to maintain a predefined
MaxNewSize parameters control the new generation's minimum and maximum size. Regulate the new generation size by setting these parameters equal. The bigger the younger generation, the less often minor collections occur. The size of the young generation relative to the old generation is controlled by
NewRatio. For example, setting
-XX:NewRatio=3 means that the ratio between the old and young generation is 1:3, the combined size of eden and the survivor spaces will be fourth of the heap.
By default, the GlassFish Server is invoked with the Java HotSpot Server JVM. The default
NewRatio for the Server JVM is 2: the old generation occupies 2/3 of the heap while the new generation occupies 1/3. The larger new generation can accommodate many more short-lived objects, decreasing the need for slow major collections. The old generation is still sufficiently large enough to hold many long-lived objects.
To size the Java heap:
Decide the total amount of memory you can afford for the JVM. Accordingly, graph your own performance metric against young generation sizes to find the best setting.
Make plenty of memory available to the young generation. The default is calculated from
NewRatio and the
Larger eden or younger generation spaces increase the spacing between full GCs. But young space collections could take a proportionally longer time. In general, keep the eden size between one fourth and one third the maximum heap size. The old generation must be larger than the new generation.
For up-to-date defaults, see Java HotSpot VM Options.
This is an exmple heap configuration used by GlassFish Server on Solaris for large applications:
-Xms3584m -Xmx3584m -verbose:gc -Dsun.rmi.dgc.client.gcInterval=3600000
SurvivorRatio parameter controls the size of the two survivor spaces. For example,
-XX:SurvivorRatio=6 sets the ratio between each survivor space and eden to be 1:6, each survivor space will be one eighth of the young generation. The default for Solaris is 32. If survivor spaces are too small, copying collection overflows directly into the old generation. If survivor spaces are too large, they will be empty. At each GC, the JVM determines the number of times an object can be copied before it is tenured, called the tenure threshold. This threshold is chosen to keep the survivor space half full.
Use the option
-XX:+PrintTenuringDistribution to show the threshold and ages of the objects in the new generation. It is useful for observing the lifetime distribution of an application.
When the JVM initializes, it tries to allocate its heap using the
-Xms setting. The base addresses of GlassFish Server DLLs can restrict the amount of contiguous address space available, causing JVM initialization to fail. The amount of contiguous address space available for Java memory varies depending on the base addresses assigned to the DLLs. You can increase the amount of contiguous address space available by rebasing the GlassFish Server DLLs.
To prevent load address collisions, set preferred base addresses with the rebase utilty that comes with Visual Studio and the Platform SDK. Use the rebase utility to reassign the base addresses of the GlassFish Server DLLs to prevent relocations at load time and increase the available process memory for the Java heap.
There are a few GlassFish Server DLLs that have non-default base addresses that can cause collisions. For example:
nspr libraries have a preferred address of 0x30000000.
icu libraries have the address of 0x4A?00000.
Move these libraries near the system DLLs (
msvcrt.dll is at
0x78000000) to increase the available maximum contiguous address space substantially. Since rebasing can be done on any DLL, rebase to the DLLs after installing the GlassFish Server.
To perform rebasing, you need:
Visual Studio and the Microsoft Framework SDK rebase utility
bin the default directory.
Enter this command:
rebase -b 0x6000000 *.dll
dependencywalker utility to make sure the DLLs were rebased correctly.
For more information, see the Dependency Walker website.
Increase the size for the Java heap, and set the JVM Option accordingly on the JVM Settings page in the Admin Console.
Restart the GlassFish Server.
This is an example heap configuration used by Oracle GlassFish Server for heavy server-centric applications, on Windows, as set in the
<jvm-options> -Xms1400m </jvm-options> <jvm-options> -Xmx1400m </jvm-options>
For more information on rebasing, see MSDN documentation for rebase utility.
For more information on tuning the JVM, see: