I little while ago Alex Clemm and I published a draft called "Token Cell
Routing Data Plane Concepts”
https://datatracker.ietf.org/doc/draft-bcx-rtgwg-tcr
<https://datatracker.ietf.org/doc/draft-bcx-rtgwg-tcr> which I hope we can
explain to the the RTG WG at IETF 111.
Currently the fundamental basis of packet design is linear parsing (linear as
in the media use of the term), and our observation is that a move away from
this introduces a powerful set of opportunities to the network layer, just as
the media industry found when they enabled non-linear access to content.
Normally a network layer packet is designed on the assumption that the parser
starts at the first byte of the header and looks at all of it regardless of
whether all of it is intended for processing by this node. That is what make
option parsing particularly difficult since there is no easy way of knowing if
the packet contains an option that is of interest to a node without that node
first looking at all preceding options and making a decision for itself as to
the applicability of the option to its packet processing actions. It seems
strange to me that we wish to specify at source which nodes a packet will pass
through, but require the node to determine which options, and importantly which
instance of an option applies to it.
We know BTW that just the formal network layer is in itself insufficient to
describe the required actions on the packet. Routers cheat and look outside the
network layer to glean more information of which the use of the five tuple for
ECMP is the most well known example. It is also an example where this cheating
sometimes trips them up.
In this concept proposal we looked at what we can do if we introduce pointers
to the network layer packet structure and were quite surprised at the power
that such mechanisms unleashed. This was particularly the case with latency
based forwarding, FRR, OAM, and packet disposition.
The essence of the proposal is action, parameter, pointer to next action. This
is a type of active networking, but one which is constrained since, like MPLS
the only actions are ones that were previously enabled by the operator at the
node. In other words this is the MPLS security model which we know works, but
one applied in a generalised way.
A characteristic of the design is that it is easily made address family
independent, and thus is able to support any address family provided or course
that the FIB is set up accordingly. This has some interesting advantages in
constrained networks that hopefully we can explore in more depth at a later
time.
Although we have written this up as a network layer in its own right, the
concept can be used as structure to provide meta or auxiliary data for an
existing network layer. I hope to provide more details on that in the future.
Comments on the draft are welcome, and hopefully the chairs will find time on
the agenda to explore with the WG some of the concepts so as to make future
email discussions productive.
Best regards
Stewart
_______________________________________________
rtgwg mailing list
[email protected]
https://www.ietf.org/mailman/listinfo/rtgwg
