They came for your servers twenty years ago and people thought it was a passing fad, said that it would never work. But they went ahead and swallowed most of them.
Now they are back and they want your network switches.
Who are they?
The Disaggregators — and they will break up anything they can get their hands on. They say they want you to have freedom and choice. But freedom is rarely free and choice can makes things overly complex. Fortunately, to have your cake and eat it too, there’s the big red easy button solution called Apstra.
Brief History Lesson
But let’s step back for a second. To disaggregate something in the infrastructure world we all know and love is to separate out its components so that you have the option to use different sources at each layer. Once you get hardware disaggregated, then everything after that is virtualized (i.e. it’s just 1’s and 0’s). Microsoft started the desktop and server disaggregation craze with Windows, then VMWare started the virtualization era in the 90’s and eventually had a few buddies doing similar things, e.g. KVM and Docker.
However, the network just sat around in Black Box mode where vendor’s proprietary software and hardware was fully bundled and never the two shall part. But after a few fits and starts in the early 2010’s, the full disaggregation of network hardware and network OS’s is officially ON and everybody and their grandma is coming out with Network OS’s completely disaggregated from any hardware component. And this is a very good thing because nothing spurs innovation more than competition!
Cumulus and Pica8 started the commercial offering in the early 2010’s. Sidebar: I remember “competing” with Cumulus circa 2012 at a large Financial when I was at Arista and I was so NOT at all concerned for Arista (nor Cisco) as it related to white box. Arista could barely unseat Cisco’s decades long near-monopoly of the Enterprise data center space — and remember Arista’s EOS look-and-feel is “awfully similar” to Cisco’s IOS. But suddenly, people expected The Certified, i.e. CCNA/NP/IE’s, to use config files in a Linux OS to make network changes and troubleshoot? Never gonna happen! Or so I thought.
The new offerings kept coming. SnapRoute was all the rage at one point, also a commercial offering, but haven’t heard much about them of late. Microsoft’s SONiC, Dell’s OpenSwitch (OPX) are actually FREE offerings, with both companies having different reasons for offering it up (Microsoft to spur innovation for Azure and Dell to sell more servers and own entire stack). Facebook makes their own NOS called FBOSS and is “giving it back” to OCP (so people can copy it). Amazon does whatever the hell they want to do, pretty sure they leveraged a bit of a Cumulus offering at one point but it’s a Frankenstein creation that they aren’t sharing with anyone. AT&T is building their own. Then there are the long time entrants Metaswitch and IP Infusion which continue to evolve. But what’s probably most amazing is that Cisco, Juniper, Mellanox and Arista all have their own disaggregated solutions. Although truth be told, most of them tend to keep it on the downlow and may call it something else — and are likely are only doing it so Big Cloud will continue talking to them. One mustn’t upset your Black Box derived, Wall Street analyzed revenues!
What took so long for network to get on board the disaggregation train? Well when you have a near monopoly in the Enterprise, like Cisco did, the only goal was to sell more boxes so the focus was on improving “speeds and feeds”. But then Arista showed up and Enterprises had a viable choice. In parallel, the Cloud started it’s march towards maturity, so now Enterprise IT Departments had competition. If they didn’t get their act together, the Business was going to just use the Cloud and executives would question why they even had an Enterprise IT department. Companies also tired of vendor lock-in across all technology silos. Oh, and perhaps the final piece of the puzzle was API’s explosion to ubiquity.
So can/would you disaggregate your network such that you use best of breed/best of wallet decision making and winding up with Vendor A in the spine, Vendor B for HPC, Vendor C for VDI, Vendor D for Low Latency App, etc? That sounds like an Operational nightmare!
The Humpty Dumpty Problem
Now that everybody has disaggregated everything into separate pieces, who is going to put Humpty Dumpty back together again? Well, Apstra can not only accomplish this but it also provides support for the combinations that are part of their solution, so you can be completely de-risked in your ability to take advantage of your choices, including any Open Source, formerly unsupported offerings.
What’s more, you could still have the Black Box solutions in parallel as you get more and more comfortable with “NOS of the day”. And if you don’t like one implementation for either a technical or business reasons, click a button and Apstra swaps it for another in seconds — how’s that for a PoC?
Disaggregation is only ready for prime time if there is a supported means to reaggregate. So give me your tired, your poor, your huddled masses of Network OS’s yearning to breathe free… and Apstra will vet them and guarantee support for the ones that produce as advertised!