yes, agreed. I probably should have mentioned that aspect :)
> So don't fall into the trap of thinking that zero-copy is always
> obviously a win. It isn't.
oh, I don't think anyone actually hacking on this stuff feels that way.
Donald was not entirely excited about talking with us about this stuff
becuase I fear he got the impression that we intend to add an overly
abstracted slow as snot general mechanism and bog down the currently fast
paths. Heavens no.
But I still think that doing dma/csum from non-skb owned pages can be a
very good thing in certain circumstances. I'm specifically interested in
high bandwidth transmit. receive is a nightmare with the existing API,
but with sendfile() or from kernel sunrpc land we should be able to keep
the overhead down enough to make freeing the cpu from touching all the
outgoing data worth while.
we're _certainly_ not advocating replacing the entire skb scheme with
mbufs and unconditionally doing page flipping zero copy on rx, etc, etc,
etc.
-- zach
- - - - - -
007 373 5963
-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@vger.rutgers.edu
Please read the FAQ at http://www.tux.org/lkml/