There's two ways to create a ByteBuffer, where one can be subdivided again.

If you have an already existing byte[], you can "wrap" it into a ByteBuffer to simplify processing:

byte[] reqBuffer = new byte[BUFFER_SIZE];
int readBytes = socketInputStream.read(reqBuffer);
final ByteBuffer reqBufferWrapper = ByteBuffer.wrap(reqBuffer);

This would be a possibility for code that handles low-level networking interactions

---

If you do not have an already existing byte[], you can create a ByteBuffer over an array that's specifically allocated for the buffer like this:

final ByteBuffer respBuffer = ByteBuffer.allocate(RESPONSE_BUFFER_SIZE);
putResponseData(respBuffer);
socketOutputStream.write(respBuffer.array());

If the code-path is extremely performance critical and you need direct system memory access, the ByteBuffer can even allocate direct buffers using #allocateDirect()

Given a ByteBuffer instance one can write primitive-type data to it using relative and absolute put. The striking difference is that putting data using the relative method keeps track of the index the data is inserted at for you, while the absolute method always requires giving an index to put the data at.

Both methods allow "chaining" calls. Given a sufficiently sized buffer one can accordingly do the following:

buffer.putInt(0xCAFEBABE).putChar('c').putFloat(0.25).putLong(0xDEADBEEFCAFEBABE);

which is equivalent to:

buffer.putInt(0xCAFEBABE);
buffer.putChar('c');
buffer.putFloat(0.25);
buffer.putLong(0xDEADBEEFCAFEBABE);

Do note that the method operating on bytes is not named specially. Additionally note that it's also valid to pass both a ByteBuffer and a byte[] to put. Other than that, all primitive types have specialized put-methods.

An additional note: The index given when using absolute put* is always counted in bytes.

DirectByteBuffer is special implementation of ByteBuffer that has no byte[] laying underneath.

We can allocate such ByteBuffer by calling:

ByteBuffer directBuffer = ByteBuffer.allocateDirect(16);

This operation will allocate 16 bytes of memory. The contents of direct buffers may reside outside of the normal garbage-collected heap.

We can verify whether ByteBuffer is direct by calling:

directBuffer.isDirect(); // true

The main characteristics of DirectByteBuffer is that JVM will try to natively work on allocated memory without any additional buffering so operations performed on it may be faster then those performed on ByteBuffers with arrays lying underneath.

It is recomended to use DirectByteBuffer with heavy IO operations that rely on speed of execution, like real time communication.

We have to be aware that if we try using array() method we will get UnsupportedOperationException. So it is a good practice to chech whether our ByteBuffer has it (byte array) before we try to access it:

 byte[] arrayOfBytes;
 if(buffer.hasArray()) {
     arrayOfBytes = buffer.array();
 }

Another use of direct byte buffer is interop through JNI. Since a direct byte buffer does not use a byte[], but an actual block of memory, it is possible to access that memory directly through a pointer in native code. This can save a bit of trouble and overhead on marshalling between the Java and native representation of data.

The JNI interface defines several functions to handle direct byte buffers: NIO Support.