There is something rather nice going on in networks.

If you have (like me) mostly ignored the blizzard of buzzwords, not ready to invest the time to sort hype from substance, this is missive is meant as a shortcut.

(Skip the links I am offering in the middle, at first. I promise to tie this together at the end.)

The acronym SDN (software defined networking) could mean almost anything. This means that vendors and pundits have spun tales to serve their own interests, and all that writing makes for confusion.

There is a central thread. Once you find the thread, everything makes sense.

Several years back, some folk wanted to do clever things with networks, and proposed a clever factoring. Networks have a certain amount of intelligence, in how packets are routed. The point at which packets are routed can be called either switches or routers. Either term refers to the intelligence of the network. The clever folk took existing switches as example. They partitioned the problem of switching into a simple/fast per-packet (often hardware based) rules engine, and an arbitrarily complex policy engine (much less often used) that could do … almost anything. Turns out they choose very well.

The central thread is OpenFlow which spins out of a research project at Stanford.

As they chose the right factoring, the right API between hardware switching and software … everything changed.

Turns out Google converted one of the largest planet-scale networks over to OpenFlow with good results, as publicly noted in April of 2012 (and PDF).

Existing switches could readily offer OpenFlow APIs, and performance of OpenFlow switches could be optimized.

This natural shift in technology is a threat to the existance of network switch vendors, as reflected in the slightly manic postings from Cisco.

Most relevant to virtualization, the Open vSwitch code is a software OpenFlow switch very well supported in Linux (with strong backing from Redhat). In cloud environments we have a great need for good support of intelligent switching, to support complex network topologies, and a fraction will be pure software switches within the hypervisor host.

Where do we end?

If you use networks, this is very good news. Your networks are going to get smarter, cheaper, faster, and easier to use.

Expect a lot of noise as existing vendors try to defend their old model. Likely you will deploy more switches, when they become a lot easier to use, and cheaper. Naturally, the existing switch vendors will try to convince you to keep buying their proprietary gear - with steadily less reason. The core is OpenFlow, and supporting products (hardware and software).

Very cool.

Update - 2015-03-30: Looks like Broadcom is building a large new facility in Irvine, along my present commute. Have to wonder if this is driven in part by the rise of “merchant silicon” enabled by OpenFlow.