FileChannel和FileInputStream中read（ByteBuffer）和read（byte []）之间的区别

Question

I am newly to NIO and I find an article saying ' the block-based transmission is commonly more effective than stream-based transmission '. It means read(ByteBuffer) is block-based transmission and read(byte[]) is stream-based transmission.

I want to know what's the internal difference between the two methods.

ps:I also hear block-based transmission is transferring byte arrays and stream-based transmission is transferring byte one by one. I think it's wrong, because java.io.FileInputStream.read(byte[]) transfers byte array as well.

Answer 1

One thing that makes Bytebuffer more efficient is using direct memory. This avoids a copy from direct memory into a byte[]. If you are merely copying data from one Channel to another this can be up to 30% faster. If you reading byte by byte it can be slightly slower to use a ByteBuffer as it has more overhead accessing each byte. If you use it to read binary eg int or double it can be much faster as it can grab the whole value in one access.

Answer 2

I think you're talking about buffer-based vs stream-based I/O operations. Java NIO is buffer oriented in the sense that data is first read into a buffer which is then processed. This gives one flexibility. Also, you need to be sure that the buffer has all the data you require, before you process it. On the other hand, with stream-based I/O, you read one or more bytes from the stream, these are not cached anywhere. This is a blocking I/O, while buffer-based I/O (which is Java NIO) is a non-blocking IO.

Answer 3

While I wouldn't use "stream-based" to characterize read(byte[]) , there are efficiency gains to a ByteBuffer over a byte[] in some cases.

See A simple rule of when I should use direct buffers with Java NIO for network I/O? and ByteBuffer.allocate() vs. ByteBuffer.allocateDirect()

The memory backing a ByteBuffer can be (if "direct") easier for the JVM to pass to the OS and to do IO tricks with (for example passing the memory directly to read to write calls), and may not be "on" the JVM's heap. The memory backing a byte[] is on the JVM heap and IO generally does not go directly into the memory used by the array (instead it often goes through a bounce buffer --- because the GC may "move" array objects around in memory while IO is pending or the array memory may not be contiguous).

However, if you have to manipulate the data in Java, a ByteBuffer may not make much difference, as you'll eventually have to copy the data into the Java heap to manipulate it. If you're doing a data copy in and back out with out manipulation, a direct ByteBuffer can be a win.