The control plane, removed

A subject the papers are about. The loosest grouping, and the one to reach for last.

What happens to a network when the control plane is lifted out of the boxes that forward packets. The set runs from the argument that policy and mechanism were tangled in the first place, through the state that a switch then no longer has to keep, to what a single programmable controller is worth as a target.

OpenFlow is the hinge and everything sorts around it. Before it: Livermore on controlling an adaptive flat network in 1998, and the MPLS control-plane security report, which is the previous generation's version of the same worry. Griffioen states the separation argument directly. After it, Kreutz's survey is the standard orientation and Hari is the concrete payoff -- path switching, where knowing the path lets a switch stop keeping per-flow state, which is the set's title as an engineering result rather than an architecture diagram.

The deployments are the part most worth reading and the part most often skipped. Faucet is a production controller; Gupta's industrial-scale software-defined exchange and Bruyere's rethinking of the IXP are the same idea applied where it pays best. Those two connect this set to the IXP thread in `who-the-internet-depends-on`: an exchange point is where centralised control has the clearest business case, because the policy being expressed is between parties rather than inside one.

Security is the largest single thread here, and that follows from the premise rather than being a separate topic. Lakshminarayanan, the Scott-Hayward survey, Black on mitigations and the NCSC guidance all exist because moving control into one programmable place creates something that was not previously there to attack.

Two records answer the question differently. SCION does not centralise control; it makes the path explicit to the sender, which is the path-aware alternative to a controller. Vemuganti carries source routing into LEO satellite constellations, where a central controller is least plausible.

Hall's *Collaborating with the enemy on network management* has a companion transcript-of-discussion record in the corpus that is not in this set.