On 1/30/2014 12:20 PM, Scott Brim wrote:
There are two topics that should be separate: checksums and congestion
control.
Agreed.
On Thu, Jan 30, 2014 at 2:28 PM, Joe Touch <touch(_at_)isi(_dot_)edu
<mailto:touch(_at_)isi(_dot_)edu>> wrote:
I think we agree that if the router needs faster, then the router
gets faster.
That's in response to Ross, about checksums.
Also about the complaint that routers run to fast to run any sort of
congestion algorithm, including circuit breakers (which is clearly false).
Where we disagree is whether the "bull in the china shop" gets to
ignore the rules of the road to get what it wants.
congestion
Yes, in the sense that incorrect checksums should be checked at the
egress only anyway, and thus ignoring them would only impact the network
after the egress.
No, in the sense that anything that peeks inside a UDP header has the
right to want to check whether that data is uncorrupted - that includes
things like NATs.
I.e., I won't disagree on what router vendors want or what they're
willing to support (though I will disagree on what hardware can do,
based on experience).
checksums
Both, as per above.
I will disagree that routers get to run roughshod over the rest of
us to do it.
Simple solution -- don't use UDP.
congestion.
Both. If you want the benefits of UDP, pay the penalty. If you don't
want the penalty, don't use UDP.
I don't buy the claims that there's a need here to eat cake and have it too.
Joe, do I understand correctly that you've conceded about checksums and
now you want to talk about congestion again?
I'm talking about the overall argument, which has circled back to "we
need to do what vendors are willing to do", rather than "vendors need to
play nice in the Internet that they're making money off of".
Joe