On 1/14/2014 4:40 PM, Curtis Villamizar wrote:
In message <DB6CF60F-FFBA-47DA-9FD6-7288CCB260A6(_at_)netapp(_dot_)com>
"Eggert, Lars" writes:
On 2014-1-14, at 15:20, Stewart Bryant <stbryant(_at_)cisco(_dot_)com> wrote:
Yes, the inner (real) transport header is the only meaningful place
to apply congestion avoidance.
But what if the inner traffic isn't congestion controlled?
Lars
Lars,
The exact same thing will happen in all of the following cases:
NON-congestion controlled application --over--
UDP --over-- IP --over-- L2
NON-congestion controlled application --over--
UDP --over-- IP --over-- MPLS --over-- L2
NON-congestion controlled application --over--
UDP --over-- IP --over-- MPLS --over-- UDP --over-- IP --over-- L2
The non-congestion controlled application is what needs fixing.
In all three cases, RFC5405 expects that traffic inside UDP is
congestion controlled. That can happen when the source application does
so, but when that isn't known, it needs to happen at whatever layer puts
the packet inside the UDP header that the Internet ends up seeing.
It would be wrong to try to put congestion control at every layer
underneath the non-congestion controlled application
Yes, but it's not wrong to require some sort of congestion control at
any layer that generates a UDP packet inside the Internet or any other
Internet-protocol-based network that has the same RFC5405 expectations.
Joe