Networks are changing. What was once limited to hardware specialists at a handful of vendors is now becoming accessible to the rest of us. And by “us” I mean system builders, software developers, and infrastructure operators.
I’ll explain by contrast. Until very recently, networking has followed a hardware-centric design and supply model. The vendor who supplies the gear also builds the ASIC(s), designs the board(s) and the packaging, writes the software, and exposes a proprietary control interface. Closed, vertically integrated, and slow to change, that was networking (and largely still is).
Over the last few years, however, I’ve been involved in multiple efforts in which non-vendors have built their own systems. The model varies. Some build their own switches through direct relationship with contract manufacturers, some shortcut the vendors by buying whiteboxes directly from the ODM and write their own switch software, and others use traditional networking equipment but pull functionality into soft switches at the network edge. The point is, they’re doing it themselves, and building networks far cheaper, far more innovative, and generally for more awesome than are available from the vendors.
Why are they doing this? Because modern deployment environments demand better networks: the scale of big data, the dynamics of compute virtualization, rapid provisioning requirements for IT, the cost model of cloud computing, and the list goes on.
The broader implications of this trend are still in question. However, history has shown that when there is a paradigm shift in the industry, the leaders of the old model rarely retain their positions.
What is certain, is that going forward, vendors and users alike will have to build networks differently. Very differently.
And that is what this blog is about. It’s about programmable networks. It’s about soft switching and network virtualization. It’s about SDN/OpenFlow, and the trend towards merchant silicon. It’s about building networks as distributed systems. But mostly, it’s about the overdue demise of the vertically integrated network model.