[ncrg] SDN addresses control plane complexity, not data plane

[Bob Briscoe <[e-mail]>]

Dave,

At one point in your talk on SDN & complexity, you pointed to just 
under the neck of the classic TCP/IP hourglass as the point that the 
originators of SDN chose to address.

The message wrt complexity is really the reverse. SDN doesn't attack 
the narrowest (least diverse) layer, it attacks the fattest (most 
diverse) layer, in an attempt to narrow it to a single standard 
(cutting out complexity).

Reason: The hour-glass represents the data plane. SDN concerns the 
control plane (specifically routing). The equivalent of the 
hour-glass in the control plane has evolved into an inverse 
hour-glass - a fatty-glass. See slide #8 or even #15 in Rui's 
presentation here:
<https://www.net.t-labs.tu-berlin.de/arcadia/talks/Aguiar.pdf>
or the paper: "Some comments on hour glasses"
<http://dl.acm.org/citation.cfm?id=1452346>

The more important a layer is for interoperability, the thinner the 
layer. All the different routing protocols make the control plane 
into a fatty-glass, because there is no /operational/ need for 
interop between routing protocols within different ASs.

SDN attempts to narrow the control fatty-glass by standardisation of 
intra-domain policy-routing at build time, which implies the industry 
is maturing from a focus on run-time conformity (between operators) 
in the data plane by adding build-time conformity (between vendors) 
in the control plane.


Bob



________________________________________________________________
Bob Briscoe,                                                  BT