A construct that has
not been mentioned, but is useful in this discussion is mp2p.
Based on Neil's
definition, mp2p is not a connection. This makes sense when you think about
it. At the reciever, additional information is needed to identify the sender
of each packet in order to recover the individual streams. So, mp2p is a
multiplexing action which requires demultiplexing at the
receiver.
This is useful to keep
in mind when talking about mp2mp. A mp2mp construct is actually a combination
of p2mp and mp2p.
From the point of view
of the sender, mp2mp looks like a p2mp connection. In the MPLS world, any
packet entering the network will take a fixed and known path to the recievers.
Looking at only this behavior, mp2mp looks like a connection.
From the point of view
of the receiver, mp2mp looks like mp2p. The receiver requires additional
information in order to identify and recover the different traffic
streams.
John
An observation. There is no such construct as a mp2mp
*connection*.....one can have any-any in a cl-ps mode network (precisely
because there are no connections) but it is NOT possible to have mp2mp
connections in either a co-ps or co-cs mode network. This can be proved
by an appeal to (Shannon) information theory.....but it would require a very
lengthy explanation, especially wrt how labelling works (its not as simple as
some folks might assume).
However, let me simply state here what the bottom-line
conclusions/rules are:
A connection must:
-
only have a single source
- only have connectivity of 1
....so only p2p and p2mp connection
constructions are allowed.
Further, and consequential from the above:
-
ordering of traffic units is preserved in a connection
- once created there is no routing choice available in a
connection
A trail is 'the source to a specific-sink'
relationship in these connection
constructs.
regards, Neil
Some
critical notes wrt mp2mp LSP:
The
terms upstream and downstream relate to the terms ingress and egress, but
not to root (and
three times not to root while keeping open whether the root is an
ingress(+egress) at all).
The
mp2mp LSP is supposed to be a mesh-less delivery channel which interconnects
a specific set of leaf nodes such that at any junction traffic is forwarded
to all other directions except the one from where it is received. This is my
understanding about the mp2mp stream direction. The presented design with
its upstreams and downstreams all over the place is fairly a sign of
ossification, isn't it ?
Heiner
Hummel