Software Defined Networks

Can someone explain what is all the fuss? SDN is like the latest telecom craze but the articles do a poor job of explaining the advantages. I seek concrete examples.



You can start by taking a look at Openflow which embraces the SDN concept.

Ask 5 experts and you will get at least 8 definitions.

It’s a very loosely defined buzzword that encompasses a variety of technologies.

It’s the latest ill-defined term following cloud computing.


“SDN” is a broad brush.

Here’s a “concrete example” of using software to do interesting network tricks. I’m biased, of course.


SDN allows non-software guys to take down your whole network because they forgot some whitespace in a yaml file. :stuck_out_tongue_winking_eye:

It's called the "cycle of reincarnation".

Way back when, a "router" was a Microvax-II with 2 network cards in it, and everything,
down to the packet checksums, was done in software on the Microvax CPU.

Then we got "routers" that did a lot of the stuff like checksumming in hardware.

Then we went back to software for more advanced features, then the hardware got
smart enough to do it.

For a while, routers were doing IPv4 support in hardware, and the occasional IPv6
packet got tossed towards the CPU. (That was, of course, once they got smart enough
to do something other than "compare first 4 bits == "0100", drop on not-equal". I think
a few boxes didn't even check the first 4 bits and assumed that all packets started
that way, and hilarity and hijinks ensued....

Lather, rinse, repeat multiple times over the past 4 decades.

And now we're seeing "SDN" which just means "Now that the hardware is smarter
and doing a lot of the stuff we used to do in software, the CPU has more free
capacity and we can do new clever stuff in software that we couldn't do

It's *NOT TRULY* a software-defined network. If it was, there wouldn't be any
hardware support for checksumming etc, because the checksum to use would be
done in software so it could be easily replaced if you had reason to use a
different checksum algorithm.

(Hint - it's as big a crock as "Software Defined Storage" - which just means
that it's software doing things like the RAID, erasure encoding, and logical
volumes rather than physical volumes, even though the logical volumes are
usually really just RAID-0 concatenations of segments of physical volumes.
Meanwhile, "software defined radio" really means "the physical hardware is
flexible, and software is used to configure it in case you didn't want the standard
channel frequencies in the 2.4 and 5ghz bands".

SDN originally meant ‘separate the forwarding plane from the control plane, and do wacky stuff with the CP’.

It’s been applied over the years as a nice buzzwordy marketing term for just about anything that involves software interacting with network hardware in any way. Not correct of course, but that’s never stopped anyone from try to sell before. :slight_smile:

Can someone explain what is all the fuss? SDN is like the latest telecom craze but the articles do a poor job of explaining the advantages. I seek concrete examples.

Maybe our example from my <; company can illustrate. It is the size of two PhD projects in our small research institute spun of as a neighbourhood ISP and smart microgrid.

We needed a way to build our wide area network (an internet access provider) AND a supercomputer interconnect to both allow for loops in the network. None of the standard switches like Cisco and Juniper can handle this gracefully.

SDN/Openflow, especially P4, allows us to write a few programs in a few weeks that generates the hardware and software for the 120 Gbps and 6 terabit mesh switches for whatever topology of our network. We can buy white label switches for $5000 or our own $800 12x10G FPGA fabric switches or $1000 260x50 Gbps switches. We save a few million in an fiber ISP metro network to 20.000 households with 4x10 Gbps ports. More important, we don’t have to fight (adapt and patch) Cicso and Juniper legacy software and protocols every week for 10 years on switches costing 20 times as much.
We can test all our software before we built in mininet in an afternoon on a laptop. We could build a test network with OpenVswitch in on fast off the shelf linux servers with 10G ports.
We found the Slimfly topology to be $2000 cheaper per household than all other switch topologies.
My ISP WAN network winds up being almost as fast as Cray + AMD new Rome zen 2 supercomputer cluster switch fabric with the investment of three people while saving $2000 per location and a few million in my core datacenters with our own hardware product (similar to a NetFPGA Sume with 2000 in use). We moved an academic network out of the lab into a fiber ISP Wan for less than $200.000 while saving $40,000,000. A side effect of building arbitrary topologies is an additional saving of $4,000,000 in cable lengths in a 4km2 naighmerhood fiber rollout.


Merik Voswinkel,
Fiberhood Coop
Metamorph research institute

SDN is definitely an overloaded and confusing term that is used inconsistently. Here are a few attempts to explain:

— Jen

I tell everyone we had SDNs in the 90s.

But we called it “expect scripts”.


I miss TCL ...

There is also a trend for "AI" in SDNs and even cognitive networks, can drop this one here:

It's not gone. My favorite DVCS - - is written in TCL.

Miles Fidelman



You can start by taking a look at Openflow which embraces the SDN concept.

Please do not

So does anyone still wonder why we have so few women in our field?

Real nice, Töma.

@moderators, can we have a comment from you please about the acceptability of the language used here?



Personally, the way I look to SDN today, as a use case. Saying just “SDN” is an unfair word. While it simply started as someone did not want to buy an expensive router while he can write a script on a server (controller) to program a switch to do what he wants, it drafted alot from this view.

Although the hardware got smarter and with some vendors even cheaper, still it will not replace the flexibility of software, quick integration of smart AI controller feedback, flow granularity decisions and in some cases replacement of conventional protocols…etc Not as easy as it seems and comes with compromises



Cognitive networks? This is where Skynet comes from, right?

Jen missed another ACM article to which she contributed:

- "A Purpose-built Global Network: Google's Move to SDN" (ACM Queue,
December 11, 2015, Volume 13, issue 8)
A discussion with Amin Vahdat, David Clark, and Jennifer Rexford


So like leaving the HTTP service up on router/switches as a Upper Manager trap =D.