Link Details

Even if there is a native method... it doesn't mean that it's necessarily used. Like there are native methods for lots of Math stuff... those /aren't/ used. The VM (well, Sun's VM) inlines those calls directly.

The author also forgot to include native numbers for comparison. His test doesn't run all that fast, but there is no indication that it could run faster than that.

The Java compiler/JVM cannot inline native code. That would require inlining the readBytes() behavior - not feasible.

Inlining Math operations is quite different. (You should have got the "why" of this in school.)

What about using NIO?

Benchmarking NIO is another exercise. There is no reason for FileInputStream - with large buffers - to be radically slower than C/C++ using stdio.

I think there was a good thread some time back about file streams in Java. Anyway, have you tried using BufferedInputStream?

Using BufferedInputStream would make this benchmark *slower*. If you write code for a living, you should understand why.

Here is the JNI documentation for SetByteArrayRegion http://java.sun.com/javase/6/docs/technotes/guides/jni/spec/functions.html#wp22933

I can't help but wonder why GetByteArrayElements() is not being used. You shouldn't have to allocate a buffer at all since you already have the byte array.