| Sun ONE Application Server 7 Performance Tuning Guide |
Chapter 5
Tuning the Java Runtime SystemThe Solaris operating environment, by default, supports a two level thread model (up to Solaris 8). Application level Java threads are mapped to user level Solaris threads, which are multiplexed on a limited pool of light weight processes (LWPS). Often, we need only as many LWPS as there are processors on the system, leading to conserved kernel resources and greater system efficiency. This helps when there are hundreds of user level threads. Fortunately (or unfortunately), you can choose from multiple threading models and different methods of synchronization within the model, but this varies from VM to VM. Adding to the confusion, the threads library will be transitioning from Solaris 8 to 9, eliminating many of the choices. Although we have a 2 level model, in the 1.4 VM, we have an effectively 1-to-1 thread/lwp model since the VM used LWP based sync by default.
The following topics are discussed in this chapter:
Using Alternate ThreadsYou can try to load the alternate libthread.so in /usr/lib/lwp/ on Solaris 8 by changing your LD_LIBRARY_PATH to include /usr/lib/lwp before /usr/lib. Both give better throughput and system utilization for certain applications; especially those using fewer threads.
By default, the Sun ONE Application Server uses /usr/lib/lwp. You can change the default settings to not use the LWP by removing /usr/lib/lwp from the LD_LIBRARY_PATH in the startserv script, but should be avoided unless required.
For applications using many threads, /usr/lib/libthread.so is the best library to use. Of course, see using -Xconcurrentio for applications with many threads as this will not only turn on LWP based sync, the default in 1.4, but also turn off TLABS, or thread local allocation buffers, which can chew up the heap and cause premature gcs.
To further examine the threading issues on Solaris with Java, see http://java.sun.com/docs/hotspot/threads/threads.html
Managing Memory and AllocationThe efficient running of any application depends on how well memory and garbage collection are managed. The following sections provide information on optimizing memory and allocation functions:
Tuning the Garbage Collector
Garbage collection reclaims the heap space previously allocated to objects no longer needed.The process of locating and removing the dead objects can stall any application while consuming as much as 25 percent throughput.
Almost all Java Runtime Environments come with a generational object memory system and sophisticated garbage collection algorithms. A generational memory system divides the heap into a few carefully sized partitions called generations. As these objects accumulate a low memory condition occurs forcing garbage collection to take place. The efficiency of a generational memory system is based on the observation that most of the objects are short lived.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. New objects allocate in eden. Longer lived objects are moved from the new generation and tenured 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 -- copied to a 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 (GC).
The frequent young space collections are quick (few milliseconds), and the occasional full GC takes a relatively longer time (tens of milliseconds to even a few seconds, depending upon the heap size).
Other garbage collection algorithms, such as the Train algorithm, are incremental. They chop down the full GC into several incremental pieces. This provides a high probability of small garbage collection pauses even when full GC takes affect. This 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. User application level object pooling may have been more beneficial running on earlier generation Java Virtual Machines. Consider pooling only if the object construction cost is very high and shows up being significant in the execution profiles.
For detail information on Tuning Garbage Collection, see http://java.sun.com/docs/hotspot/gc/index.html
Tracing Garbage Collection
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 garbage collection.
Pauses are times when an application appears unresponsive due to garbage collection. Users can have different requirements of garbage collection. A server centric application may consider throughput to be a metric, but a short pause may upset a graphical program. There are two other considerations - Footprint and promptness.
Footprint
Footprint is the working set of a 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 sizing chooses a trade-off between these considerations. A very large young generation may maximize the throughput, but does so at the expense of footprint and promptness. Pauses can be minimized by using a small young generation and incremental collection.
Pauses due to Generation Collection are inspected by diagnostic output of the Java Virtual Machine. The command line argument "-verbose:gc" prints information at every collection. Below is a sample output of the information generated when this flag is passed to the Java Virtual Machine.
[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]The numbers before and after the arrows indicate the combined size of live objects before and after the collection. The number in the parenthesis is the total available space, which is the total heap minus one of the survivor spaces. In this sample, there are 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.
Specifying Other Garbage Collector Settings
For applications which dynamically generate and load classes, permanent generation is not relevant to the GC performance. For applications which dynamically generate and load classes (JSP's), the permanent generation is relevant to the GC performance, as filling the permanent generation can trigger a Full GC. The maximum permanent generation can be tuned with -XX:MaxPermSize option.
Applications can interact with the garbage collection by invoking collections explicitly through the System.gc() call. But, relying on the application to manage the resources is a bad idea since these force major collections, and inhibit scalability on large systems. This can be disabled by using the flag -XX:+DisableExplicitGC.
The Sun ONE Application 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. The frequency of these collections can be controlled 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.
To specify the attributes for the Java virtual machine:
Tuning the Java Heap
This section discusses topics related to tuning the Java Heap for performance.
Guidelines for Java Heap Sizing
You can control the heap size using various parameters.
The -Xms and -Xmx parameters defines the minimum and maximum heap sizes. Since collections occur 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 collection 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:MinHeapFreeRatio=<minimum> and -XX:MaxHeapFreeRatio=<maximum>; and the total size bounded by -Xms and -Xmx.
Server side applications set the values of -Xms and -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 NewRatio.
The NewSize and MaxNewSize parameters control the new generation's minimum and maximum size. You can regulate the new generation size by setting these parameters equal. The bigger the younger generation, the less often minor collections occur. By default, the young 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. In order to be safe, set the NewSize/MaxNewSize to the same value.
By default, the Sun ONE Application 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.
The following are important guidelines for sizing Java heap.
- Decide the total amount of memory you can afford to 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 for 1.4 is calculated from NewRatio and the -Xmx setting.
- Larger Eden or younger generation spaces increase the spacing between full garbage collections. But young space collections could take a proportionally longer time. In general, you could keep the eden size between 1/4th and 1/3rd the maximum heap size.
Survivor Ratio Sizing
The 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. With JDK 1.4, the Solaris default 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 garbage collection, the JVM chooses a threshold number of times an object can be copied before it is tenured.
This threshold is chosen to keep the survivors half full.
The option -XX:+PrintTenuringDistribution can be used to show the threshold and ages of the objects in the new generation. It is useful for observing the lifetime distribution of an application.
For up-to-date defaults, refer to http://java.sun.com/docs/hotspot/VMOptions.html
Sample Heap Configuration on Solaris
This is a sample heap configuration used by Sun ONE Application server for heavy server-centric applications, on Solaris, as set in the server.xml file.
<jvm-options> -Xms3584m </jvm-options>
<jvm-options> -Xmx3584m </jvm-options>
<jvm-options> -verbose:gc </jvm-options>
<jvm-options> -Dsun.rmi.dgc.client.gcInterval=3600000 </jvm-options>
Sample Heap Configuration on Windows
This is a sample heap configuration used by Sun ONE Application server for heavy server-centric applications, on Windows, as set in the server.xml file.
<jvm-options> -Xms1400m </jvm-options>
<jvm-options> -Xmx1400m </jvm-options>
HotSpot Virtual Machine Tuning Options
HotSpot is a just-in-time byte-code compiler to improve the performance of Java applications. It can be highly tuned for performance. When using the Sun ONE Application Server examine the HotSpot documentation to see if the JVM is tuned appropriately.
The following web pages are recommended:
- Java HotSpot VM Options http://java.sun.com/docs/hotspot/VMOptions.html
- Frequently Asked Questions About the Java HotSpot Virtual Machine http://java.sun.com/docs/hotspot/PerformanceFAQ.html
