Possible practical implementation of McVoy's splice() io-model f
As much as DFly is breaking new ground, I think it can't hurt to stay
informed about other kernel progress. According to latest emails on
linux-kernel, linus is starting a implementation of the splice()
io-model for linux, at the first step in the form of buffering for
pipes. The basic implementation is not to send data buffers forth and
back but to pass buffer descriptors, which are ref-counted, between
interested parties.
This is related to the ideas from Matt about explicit separation of
data and command channels for the message-bassed buffering on the DFly
SSI clusters.
Just for convenience, here are the relevant links:
lkml thread:
http://marc.theaimsgroup.com/?i=%3CPine.LNX.4.58.0501070735000.2272%20()%20ppc970%20!%20osdl%20!%20org%3E
Original proposal from Larry McVoy:
http://www.bitmover.com/lm/papers/splice.ps
--
Greetz, Antonio Vargas aka winden of network
http://wind.codepixel.com/
Las cosas no son lo que parecen, excepto cuando parecen lo que si son.