Primer on SDN (think ANET)

Short and sweet, just a little nerdy:

Talks about SDN as a technology and what it’s advantages are, nothing about Arista in particular.

1 Like

I highly recommend anyone interested in networking vendors read this paper.…

Please pay attention to this foot note on page 10.

6 N.B. Although AT&T accepts that proprietary interfaces fit the architecture described by SDN, there is a strong aversion to being locked-in to a vendor-specific protocol as it’s unlikely to allow us to reach our white box vision.

From page 13, my guess is that AT&T will avoid any vendor lock-in type of equipment. If vendor lock-in, (AKA Cisco) is seen, AT&T will avoid them. So, a company that has good hardware with open source connections into it will do well. Also, those that can stay innovative on the software front will also do well.

SDN as Disruptive Technology

It was the intention of SDN to develop a fresh approach to networking, and it’s no surprise that the SDN technology is often seen as disruptive. Some of these disruptive aspects include:
Shift in Value-Add – with the separation of packet forwarding and control, new opportunities emerge for new suppliers. Vendor lock-in is reduced, and value-add can be provided independently from providing the hardware.

Enhancing Merchant Silicon – With a sophisticated external controller, a simple network element based on merchant silicon can functionally compete with a much more sophisticated element where it could not before. Special-purpose network elements like firewalls, load distributors, and various types of gateways might become special-purpose software that can be applied to existing systems to augment their capabilities.

Instant Overlays – by controlling a distributed set of software packet switches (e.g. some type of vSwitch) and connecting them with overlay tunnels, sophisticated networks can be created and modified in near real time. Moreover, these networks often don’t need much support from underlying physical networks – those have been abstracted into a common fabric. AT&T already uses this approach in our SilverLiningTM product.

Cloud Networks – Leveraging the virtualization of network resources, new types of networks can be instantiated on shards of existing equipment. Using a single physical network element efficiently for disparate applications can allow de-fragmenting leftover capacity in existing deployments and reduce the number of physical devices in the network. Moreover, if this can be coupled with the development of a common SDN & NFV infrastructure that can be delivered in pods rather than discrete components, then the operational cost advantages can contribute toward the goal of providing networking services much more efficiently than is done today.



I was not a network engineer at my former job, but I helped manage a major transition from Big Blue mainframe architecture over 3270 protocol to a distributed UNIX architecture. One of the objectives was to avoid vendor lock-in.

We specified all bids must be POSIX compliant (a set of cross industry standards). And indeed every vendor that bid for servers was “POSIX compliant” with their own proprietary extensions. This was their way of maintaining industry compliance from a marketing perspective, but also gave them a fair degree of lock-in because every time an extension got used (which was frequently) it made it impossible to swap out the box for another one.

We told our database guys to use only vanilla SQL and stop using DB procedures, the goal was to keep the applications independent of the database. This also raised a huge outcry from the DBA community (and even some objections from the application folks), The goal was to be able to swap one DBMS for another without having to rework a bunch of applications. This didn’t fly either. In other words, avoiding vendor lock in is a myth.