#8 More efficient Rreads and Twrites

[Jon Blower]

Currently the implementations of RreadMessage and
TwriteMessage involve the copying of ByteBuffers into
byte[] arrays. Would be more efficient if we could
just keep the ByteBuffers. However, the current
implementation is reliable.

Need to understand properly the real benefits to be
gained by this more efficient system, as the code will
be more complex.

The implementations of Rread/Twrite in CVS on
31/10/2005 should contain attempts at the more
efficient system, but I found problems when trying to
write the message in two steps (one step for the
header, one for the payload). The payload sometimes
never arrived. Need to understand this better or find
an efficient way to write both at once to the network.

(When I did benchmarking for AHM2005 paper, I achieved
95% of HTTP speed with 64KB messages and 10
simultaneous read requests. Now it is down to 52%.
Clearly significant efficiencies can be made.