> There is however some validity in implimenting particular aspects on the
> NIC. Specifically the TCP checksum off loading that a number of gigabit
> NIC's have in silicon.
absolutely, as it is useful that the NIC is able to follow linked list
of arbitrarily aligned/sized blocks etc, but that is
Ron,
Didn't see you at FreeBSDcon..
Som if it's on the card how do you do things like ipfw?
And how do you route between cards?
Who keeps the routing table?
how does routed (or whatever) communicate with it?
how do you set tcp options?
you can do all these but it get's really complicated..
You
> Also there is the usual lack of flexibility in having things done
> in hw/microcode vs. software.
>
> And finally, i think having the proto in the NIC may be useful for maybe
> a router or a box doing traffic switching, but for things like a
> web server or a cache you also have to look at opti
> I'm wondering if anyone in this group has done or knows of a good profile
> of a tcp send going from user mode to bits on the wire. Reason I'm asking
> is the old "put TCP/IP on the NIC" is once again rearing its head, and I'm
> hoping there are numbers I can point to (ones that aren't old, that
> I'm wondering if anyone in this group has done or knows of a good profile
> of a tcp send going from user mode to bits on the wire. Reason I'm asking
> is the old "put TCP/IP on the NIC" is once again rearing its head, and I'm
> hoping there are numbers I can point to (ones that aren't old, that
I'm wondering if anyone in this group has done or knows of a good profile
of a tcp send going from user mode to bits on the wire. Reason I'm asking
is the old "put TCP/IP on the NIC" is once again rearing its head, and I'm
hoping there are numbers I can point to (ones that aren't old, that is)
abo
6 matches
Mail list logo
