The Artima Developer Community
Sponsored Link

Chapter 5 of Inside the Java Virtual Machine
The Java Virtual Machine
by Bill Venners

<<  Page 6 of 13  >>


The Heap

Whenever a class instance or array is created in a running Java application, the memory for the new object is allocated from a single heap. As there is only one heap inside a Java virtual machine instance, all threads share it. Because a Java application runs inside its "own" exclusive Java virtual machine instance, there is a separate heap for every individual running application. There is no way two different Java applications could trample on each other's heap data. Two different threads of the same application, however, could trample on each other's heap data. This is why you must be concerned about proper synchronization of multi-threaded access to objects (heap data) in your Java programs.

The Java virtual machine has an instruction that allocates memory on the heap for a new object, but has no instruction for freeing that memory. Just as you can't explicitly free an object in Java source code, you can't explicitly free an object in Java bytecodes. The virtual machine itself is responsible for deciding whether and when to free memory occupied by objects that are no longer referenced by the running application. Usually, a Java virtual machine implementation uses a garbage collector to manage the heap.

Garbage Collection

A garbage collector's primary function is to automatically reclaim the memory used by objects that are no longer referenced by the running application. It may also move objects as the application runs to reduce heap fragmentation.

A garbage collector is not strictly required by the Java virtual machine specification. The specification only requires that an implementation manage its own heap in some manner. For example, an implementation could simply have a fixed amount of heap space available and throw an OutOfMemory exception when that space fills up. While this implementation may not win many prizes, it does qualify as a Java virtual machine. The Java virtual machine specification does not say how much memory an implementation must make available to running programs. It does not say how an implementation must manage its heap. It says to implementation designers only that the program will be allocating memory from the heap, but not freeing it. It is up to designers to figure out how they want to deal with that fact.

No garbage collection technique is dictated by the Java virtual machine specification. Designers can use whatever techniques seem most appropriate given their goals, constraints, and talents. Because references to objects can exist in many places--Java Stacks, the heap, the method area, native method stacks--the choice of garbage collection technique heavily influences the design of an implementation's runtime data areas. Various garbage collection techniques are described in Chapter 9, "Garbage Collection."

As with the method area, the memory that makes up the heap need not be contiguous, and may be expanded and contracted as the running program progresses. An implementation's method area could, in fact, be implemented on top of its heap. In other words, when a virtual machine needs memory for a freshly loaded class, it could take that memory from the same heap on which objects reside. The same garbage collector that frees memory occupied by unreferenced objects could take care of finding and freeing (unloading) unreferenced classes. Implementations may allow users or programmers to specify an initial size for the heap, as well as a maximum and minimum size.

Object Representation

The Java virtual machine specification is silent on how objects should be represented on the heap. Object representation--an integral aspect of the overall design of the heap and garbage collector--is a decision of implementation designers

The primary data that must in some way be represented for each object is the instance variables declared in the object's class and all its superclasses. Given an object reference, the virtual machine must be able to quickly locate the instance data for the object. In addition, there must be some way to access an object's class data (stored in the method area) given a reference to the object. For this reason, the memory allocated for an object usually includes some kind of pointer into the method area.

One possible heap design divides the heap into two parts: a handle pool and an object pool. An object reference is a native pointer to a handle pool entry. A handle pool entry has two components: a pointer to instance data in the object pool and a pointer to class data in the method area. The advantage of this scheme is that it makes it easy for the virtual machine to combat heap fragmentation. When the virtual machine moves an object in the object pool, it need only update one pointer with the object's new address: the relevant pointer in the handle pool. The disadvantage of this approach is that every access to an object's instance data requires dereferencing two pointers. This approach to object representation is shown graphically in Figure 5-5. This kind of heap is demonstrated interactively by the HeapOfFish applet, described in Chapter 9, "Garbage Collection."

Figure 5-5. Splitting an object across a handle pool and object pool.

Another design makes an object reference a native pointer to a bundle of data that contains the object's instance data and a pointer to the object's class data. This approach requires dereferencing only one pointer to access an object's instance data, but makes moving objects more complicated. When the virtual machine moves an object to combat fragmentation of this kind of heap, it must update every reference to that object anywhere in the runtime data areas. This approach to object representation is shown graphically in Figure 5-6.

Figure 5-6. Keeping object data all in one place.

The virtual machine needs to get from an object reference to that object's class data for several reasons. When a running program attempts to cast an object reference to another type, the virtual machine must check to see if the type being cast to is the actual class of the referenced object or one of its supertypes. . It must perform the same kind of check when a program performs an instanceof operation. In either case, the virtual machine must look into the class data of the referenced object. When a program invokes an instance method, the virtual machine must perform dynamic binding: it must choose the method to invoke based not on the type of the reference but on the class of the object. To do this, it must once again have access to the class data given only a reference to the object.

No matter what object representation an implementation uses, it is likely that a method table is close at hand for each object. Method tables, because they speed up the invocation of instance methods, can play an important role in achieving good overall performance for a virtual machine implementation. Method tables are not required by the Java virtual machine specification and may not exist in all implementations. Implementations that have extremely low memory requirements, for instance, may not be able to afford the extra memory space method tables occupy. If an implementation does use method tables, however, an object's method table will likely be quickly accessible given just a reference to the object.

One way an implementation could connect a method table to an object reference is shown graphically in Figure 5-7. This figure shows that the pointer kept with the instance data for each object points to a special structure. The special structure has two components:

This gives the virtual machine enough information to invoke the method. The method table include pointers to data for methods declared explicitly in the object's class or inherited from superclasses. In other words, the pointers in the method table may point to methods defined in the object's class or any of its superclasses. More information on method tables is given in Chapter 8, "The Linking Model."

Figure 5-7. Keeping the method table close at hand.

If you are familiar with the inner workings of C++, you may recognize the method table as similar to the VTBL or virtual table of C++ objects. In C++, objects are represented by their instance data plus an array of pointers to any virtual functions that can be invoked on the object. This approach could also be taken by a Java virtual machine implementation. An implementation could include a copy of the method table for a class as part of the heap image for every instance of that class. This approach would consume more heap space than the approach shown in Figure 5-7, but might yield slightly better performance on a systems that enjoy large quantities of available memory.

One other kind of data that is not shown in Figures 5-5 and 5-6, but which is logically part of an object's data on the heap, is the object's lock. Each object in a Java virtual machine is associated with a lock (or mutex) that a program can use to coordinate multi-threaded access to the object. Only one thread at a time can "own" an object's lock. While a particular thread owns a particular object's lock, only that thread can access that object's instance variables. All other threads that attempt to access the object's variables have to wait until the owning thread releases the object's lock. If a thread requests a lock that is already owned by another thread, the requesting thread has to wait until the owning thread releases the lock. Once a thread owns a lock, it can request the same lock again multiple times, but then has to release the lock the same number of times before it is made available to other threads. If a thread requests a lock three times, for example, that thread will continue to own the lock until it has released it three times.

Many objects will go through their entire lifetimes without ever being locked by a thread. The data required to implement an object's lock is not needed unless the lock is actually requested by a thread. As a result, many implementations, such as the ones shown in Figure 5-5 and 5-6, may not include a pointer to "lock data" within the object itself. Such implementations must create the necessary data to represent a lock when the lock is requested for the first time. In this scheme, the virtual machine must associate the lock with the object in some indirect way, such as by placing the lock data into a search tree based on the object's address.

Along with data that implements a lock, every Java object is logically associated with data that implements a wait set. Whereas locks help threads to work independently on shared data without interfering with one another, wait sets help threads to cooperate with one another--to work together towards a common goal.

Wait sets are used in conjunction with wait and notify methods. Every class inherits from Object three "wait methods" (overloaded forms of a method named wait()) and two "notify methods" (notify() and notifyAll()). When a thread invokes a wait method on an object, the Java virtual machine suspends that thread and adds it to that object's wait set. When a thread invokes a notify method on an object, the virtual machine will at some future time wake up one or more threads from that object's wait set. As with the data that implements an object's lock, the data that implements an object's wait set is not needed unless a wait or notify method is actually invoked on the object. As a result, many implementations of the Java virtual machine may keep the wait set data separate from the actual object data. Such implementations could allocate the data needed to represent an object's wait set when a wait or notify method is first invoked on that object by the running application. For more information about locks and wait sets, see Chapter 20, "Thread Synchronization."

One last example of a type of data that may be included as part of the image of an object on the heap is any data needed by the garbage collector. The garbage collector must in some way keep track of which objects are referenced by the program. This task invariably requires data to be kept for each object on the heap. The kind of data required depends upon the garbage collection technique being used. For example, if an implementation uses a mark and sweep algorithm, it must be able to mark an object as referenced or unreferenced. For each unreferenced object, it may also need to indicate whether or not the object's finalizer has been run. As with thread locks, this data may be kept separate from the object image. Some garbage collection techniques only require this extra data while the garbage collector is actually running. A mark and sweep algorithm, for instance, could potentially use a separate bitmap for marking referenced and unreferenced objects. More detail on various garbage collection techniques, and the data that is required by each of them, is given in Chapter 9, "Garbage Collection."

In addition to data that a garbage collector uses to distinguish between reference and unreferenced objects, a garbage collector needs data to keep track of which objects on which it has already executed a finalizer. Garbage collectors must run the finalizer of any object whose class declares one before it reclaims the memory occupied by that object. The Java language specification states that a garbage collector will only execute an object's finalizer once, but allows that finalizer to "resurrect" the object: to make the object referenced again. When the object becomes unreferenced for a second time, the garbage collector must not finalize it again. Because most objects will likely not have a finalizer, and very few of those will resurrect their objects, this scenario of garbage collecting the same object twice will probably be extremely rare. As a result, the data used to keep track of objects that have already been finalized, though logically part of the data associated with an object, will likely not be part of the object representation on the heap. In most cases, garbage collectors will keep this information in a separate place. Chapter 9, "Garbage Collection," gives more information about finalization.

Array Representation

In Java, arrays are full-fledged objects. Like objects, arrays are always stored on the heap. Also like objects, implementation designers can decide how they want to represent arrays on the heap.

Arrays have a Class instance associated with their class, just like any other object. All arrays of the same dimension and type have the same class. The length of an array (or the lengths of each dimension of a multidimensional array) does not play any role in establishing the array's class. For example, an array of three ints has the same class as an array of three hundred ints. The length of an array is considered part of its instance data.

The name of an array's class has one open square bracket for each dimension plus a letter or string representing the array's type. For example, the class name for an array of ints is "[I". The class name for a three-dimensional array of bytes is "[[[B". The class name for a two-dimensional array of Objects is "[[Ljava.lang.Object". The full details of this naming convention for array classes is given in Chapter 6, "The Java Class File."

Multi-dimensional arrays are represented as arrays of arrays. A two dimensional array of ints, for example, would be represented by a one dimensional array of references to several one dimensional arrays of ints. This is shown graphically in Figure 5-8.

Figure 5-8. One possible heap representation for arrays.

The data that must be kept on the heap for each array is the array's length, the array data, and some kind of reference to the array's class data. Given a reference to an array, the virtual machine must be able to determine the array's length, to get and set its elements by index (checking to make sure the array bounds are not exceeded), and to invoke any methods declared by Object, the direct superclass of all arrays.

<<  Page 6 of 13  >>

Sponsored Links

Copyright © 1996-2019 Artima, Inc. All Rights Reserved. - Privacy Policy - Terms of Use