IEEE: Software-defined Networks

[Rob Seastrom]

Andre Kesteloot <akesteloot at gmail.com> writes:

> [[http://theinstitute.ieee.org/static/special-report-software-defined-networks]]

At work we professional cynics often make joking reference to
"Software Defiled Networks".  People like to talk about magic fairy
lands where the application signals the network as to how much
bandwidth it needs and then gets it provisioned on the fly.  The
unfortunate reality is that the developers have no idea what happens
on the other side of their API call.  Down that path lies trouble
tickets that say things like "I asked you for 100 Mbit/second and ping
is only showing 40 milliseconds.  What's wrong?".

Particularly in IEEE-land, for obvious reasons, they think of SDN in
terms of layer-2 pipes and OpenFlow.  That's the wrong way to think of
SDN.  The right way to think of it is in terms of business-as-usual
networking with database-driven configuration pushing.  For the love
of God don't separate your control plane from your data plane - yes,
they did it in SS7, but that's about a bajillion orders of magnitude
less complex than running an IP network.

It's occasionally useful to be able to make packets flow against
gravity, such as in load balancing and network monitoring
applications.  It's a great way to deploy firewall rules across your
network (and hopefully not shoot yourself in the foot in the process).
In short, automation tools with humans doing the steering.

In some ways we've been doing "SDN" for years.  BGP-based realtime
blacklists have been around for over 15 years.  SNMP (paleo-SDN if you
think of it in the right terms) is closer to 25.  Then there's the
DOCSIS OAMP interface which has been around for 15 years or so and
allows the $12/hr phone representitive to provision your cablemodem
while-u-wait.

In short, ignore the hype.

Layer 3 - It Scales.

-r