On Mon, Jan 13, 2014 at 10:51 AM, Joel M. Halpern
<jmh(_at_)joelhalpern(_dot_)com> wrote:
Lars, are you really asking that the document say:
1) That the UDP encapsulation of MPLS can only be used when the devices
performing the encapsulation and decapsulation know what protocol (above
IP, if IP is being carried; above Ethernet and IP if the frame is
Ethernet carrying IP) is being used
2) That those devices need to engage in a congestion control protocol if
the carried packets are not TCP, SCTP, or DCCP?
Those both look like excessive and difficult requirements.
I'm concerned about TCP-over-X.25 scenarios. If the lower layer
encapsulation does not know what congestion management is being used
above, and it does what would be best assuming there isn't any,
performance can truly suck.
How was this handled in other cases?
Saying UDP is a user of IP and thus must provide congestion control
would be fine if our layering were simple and UDP was Transport, but
in this case UDP is "lower layer", an encapsulation to extend the
reach of MPLS and whatever is running over it.
Which is
fine if our goal is to pretend we are telling folks not to use UDP
encapsulation of MPLS. (I had not thought that was our goal.)
In practice, I simply do not see how anyone implementing this will pay
any attention to either requirement.
Yes.
My inclination is that this is not fundamental architecture, and we
can afford to see if there's a problem before we solve it.
Scott