Hey Stack, Todd, and others, any chance you guys could take a quick
look at the last patch on hbase-5495. Although the patch is not
complete (and deserves a good cleanup and double-checks on whether it
killed all the buffer copies), it probably gives a good idea on the
overall direction. The RPC.proto is some thing that one could start
with, and then drill deeper into the other areas. I have only tried to
minimize the copies for the request/response bodies. The copy for the
header bits would probably be noise and the current patch doesn't try
to reduce copies for those..
5945 is related to 6521 (versioning), but I think 5945 is already an
improvement over what exists and if folks agree, I could spend some
extra time to get it in shape sooner (within a day or two)..