Re: zero-copy TCP fileserving

Linus Torvalds (torvalds@transmeta.com)
4 Jun 1999 07:50:06 GMT


In article <Pine.LNX.3.96.990603112124.4332B-100000@medusa.sparta.lu.se>,
Bjorn Wesen <bjorn@sparta.lu.se> wrote:
>
>At least for TCP reception, the architecture is in place:

Only for reception - that case is much simpler, and has a special case
anyway: if you can trust the sender for other reasons, you can decide to
skip the checksum even without having any hardware support - and the
packets are still going to look the same (so tcpdump etc aren't going to
be surprised).

Trusting the sender is normal in certain cases: loopback is the obvious
one, but so are internal networks with known hardware characteristics in
clusters, for example. So for reception you can often do this as a
valid optimization even without having any special hardware in place at
all.

Linus

-
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/