Instead of allocating a temporary array, performing (blocking) I/O into that
array, and then copying the data into a Java array using SetByteArrayRegion
(readBytes is an example) or doing this in reverse (writeBytes), it might
be interesting to investigate doing direct I/O to/from the Java array using
GetPrimitiveArrayCritical. In order to avoid blocking in the critical
region, non-blocking I/O could be performed. If the I/O did not succeed
immediately on the first try, we could block (using poll under Solaris)
after calling ReleasePrimitiveArrayCritical, then retry when the operation
can be done without blocking. While removing the copying cost, there is
extra cost if we have to block. For large transfers, I suspect that saving
a malloc, memcpy, and free might more than pay for itself. It is worth
investigating. Also, for platforms with small memory footprints, saving
memory may be more important than a slight performance hit.

dean.long@Eng 1998-05-02