On 10/14/2014 10:33 AM, Phillip Hallam-Baker wrote:
We have an Internet Architecture Board. But we don't have an
architecture document.
That is a question better asked of the IAB.
...
Two of the reasons there is no IETF model are the OSI model. This is
just good enough to be a substitute for an IETF model while
simultaneously demonstrating the futility of modelling.
There is however a very simple modification to the OSI model that
suddenly makes sense. The OSI model defines the layers in the
architecture. What matters in a standards context is not what happens
inside the layer, it is how that layer interfaces to other layers.
http://www.isi.edu/rna
So rather than looking at the 'Applications Layer', instead look at
the Applications interface to the Transport layer beneath. And instead
of the Transport layer, consider the interface between Transport and
Network.
If you admit that layer interfaces are what matter, then you also need
to accept that layer names are no longer meaningful. The distinctions
that defined the layers according to OSI need not be relevant.
Modelling the Internet in this fashion allows us to broaden the
definition of the Internet. At the Network layer the Internet is the
set of devices that speak IP protocol. But at the Application layer,
the Internet is the set of devices that use the Internet class of the
DNS to resolve names.
According to RNA, each layer interface requires a way to map names
within one layer to names in another. This is why ARP, BGP, DNS, and
even Google are related.
Each interface is characterized by the identifier used to mediate the
transition to the layer beneath. So there really should be a layer
between Applications and Transport because Applications use the DNS
name identifier and Transport runs on IP addresses and port numbers.
We might as well call the layer Presentation.
App layer identifiers are (IMO) service names and DNS strings.
Transport layer identifiers are port numbers and IP addresses; Internet
transport does not exist independent of the Network layer because of the
integration of the IP pseudoheader in TCP endpoint identifiers (again,
if we're using OSI layer names).
Right now that Presentation layer is encoded into BSD sockets which in
turn are hardcoded to the hosts.txt era Internet architecture.
Not since mDNS/Bonjour.
A VPN is a filter on the Network interface.
A VPN is a partial overlay.
It sits above the Network
layer and below the Transport Layer.
It sits where it sits. Some sit on L2 (L2TP). Some sit on L3 (IPsec).
Some sit at the app layer (SSH).
Software Defined Networking is a
filter on the Data Link Layer.
SDN is the current flavor of an application layer network management
protocol; IMO, nothing more, nothing less, and certainly nothing
interesting.
TLS is a bit more than a filter because it involves DNS and PKI
operations that are bundled into the sockets layer. It is really a
presentation layer.
Like TCP, TLS bundles names from a number of layers together.
Looking at the Internet as it has evolved, it fits the 'interfaces'
model really well. Building a formal model using the interfaces
approach would be fairly straightforward.
http://www.isi.edu/rna
FWIW, I'm developing this as a revision of USC's primary computer
networks course for this spring. If anyone is interested in getting some
early info and providing feedback, please contact me directly off-list.
Joe