Understanding the Java Memory Model and Garbage Collection

In this article we will try to understand the Java memory model and how garbage collection works. In this article I have used JDK8 Oracle Hot Spot 64 bit JVM. First let me depict the different memory areas available for Java processes.

Once we launch the JVM, the operating system allocates memory for the process. Here, the JVM itself is a process, and the memory allocated to that process includes the Heap, Meta Space, JIT code cache, thread stacks, and shared libraries. We call this native memory. “Native Memory” is the memory provided to the process by the operating system. How much memory the operating system allocates to the Java process depends on the operating system, processor, and the JRE. Let me explain the different memory blocks available for the JVM.

Heap Memory: JVM uses this memory to store objects. This memory is in turn split into two different areas called the “Young Generation Space” and “Tenured Space“.

Young Generation: The Young Generation or the New Space is divided into two portions called “Eden Space” and “Survivor Space".

Eden Space:When we create an object, the memory will be allocated from the Eden Space.

Survivor Space: This contains the objects that have survived from the Young garbage collection or Minor garbage collection. We have two equally divided survivor spaces called S0 and S1.

Tenured Space: The objects which reach to max tenured threshold during the minor GC or young GC, will be moved to “Tenured Space” or “Old Generation Space“.

Oracle Java Certification Exam Preparation Courses.*

*Affiliate link. See Terms of Use.

When we discuss the garbage collection process we will come to know how the above memory locations will be used.

Meta Space: This memory is out of heap memory and part of the native memory. As per the document by default the meta space doesn’t have an upper limit. In earlier versions of Java we called this “Perm Gen Space". This space is used to store the class definitions loaded by class loaders. This is designed to grow in order to avoid 0ut of memory errors. However, if it grows more than the available physical memory, then the operating system will use virtual memory. This will have an adverse effect on application performance, as swapping the data from virtual memory to physical memory and vice versa is a costly operation. We have JVM options to limit the Meta Space used by the JVM. In that case, we may get out of memory errors.

Code Cache: JVM has an interpreter to interpret the byte code and convert it into hardware dependent machine code. As part of JVM optimization, the Just In Time (JIT) compiler has been introduced. The frequently accessed code blocks will be compiled to native code by the JIT and stored it in code cache. The JIT compiled code will not be interpreted.

Now let us discuss the garbage collection process. The JVM uses a separate demon thread to do garbage collection. As we said above when the application creates the object the JVM try to get the  required memory from the eden space. The JVM performs GC as minor GC and major GC. Let us understand the minor GC.

Initially the survivor space and the tenured space is empty. When the JVM is not able to get the memory from the eden space it initiates minor GC. During minor GC, the objects which are not reachable are marked to be collected. The JVM selects one of the survivor spaces as “To Space”. It might be S0/S1. Let us say the JVM selected S0 as the “To Space”. The JVM copies the reachable objects to “To Space”, S0, and increments the reachable object's age by 1. The objects which are not fit into the survivor space will be moved to the tenured space. This process is called “premature promotion”. For this image's purpose I have made  the “To Space” bigger than the allocated space. Remember the survivor space won’t grow.

In the above diagram, the objects marked with a red color indicates that they are non-reachable. All the reachable objects are GC roots. The garbage collector won’t remove the GC roots. The garbage collector removes the non-reachable objects and empties the eden space.

For the second minor GC, the garbage collector marks the non-reachable objects from “eden space” and the “To survivor space (S0)”, copies the GC roots to the other survivor space S1, and the reachable object's age will be incremented.

In the above diagram, the objects marked with red are eligible for GC, and the other objects will be copied from the eden and survivor spaces to the other survivor space, S1, and the objects age incrementally.

The above process repeats for each minor GC. When objects reach the max age threshold, then those objects are copied to the tenured space.

There is a JVM level option called “MaxTenuringThreshold” to specify the object age threshold to promote the object to tenured space. By default the value is 15.

So it is clear that minor GC reclaims the memory from the “Young Generation Space". Minor GC is a “stop the world” process. Some times the application pause is negligible. The minor GC will be performed with single thread or multi-thread, based on the GC collector applied.

If minor GC triggers several times, eventually the “Tenured Space” will be filled up and will require more garbage collection. During this time the JVM triggers a “major GC” event. Some times we call this full GC. But, as part of full GC, the JVM reclaims the memory from “Meta Space”. If there are no objects in the heap, then the loaded classes will be removed from the meta space.

Now let us see what possibilities make the JVM trigger major GC.

• If the developer calls System.gc(), or  Runtime.getRunTime().gc() suggests the JVM to initiate GC.
• If the JVM decides there is not enough tenured space.
• During minor GC, if the JVM is not able to reclaim enough memory from the eden or survivor spaces, then a major GC may be triggered. 
• If we set a “MaxMetaspaceSize” option for the JVM and there is not enough space to load new classes, then the JVM triggers a major GC.

In upcoming articles we will analyze the garbage collection log to get insights to tune the JVM for better performance. Till then, stay tuned!