I visited the Juniper Self-Driving Network (snarky acronym BTW..) Summit yesterday in Eindhoven (The Netherlands). And it was one of those events that gave me a small glimmer of hope that there are luckily still some vendors around who are realistic about the actual adoption of automation and orchestration in the network. One speaker was clear, this process will take maybe 5 – 10 years to be fully adopted, but at the same time we can make small steps in the right direction towards a more programmable architecture. It was good to see the focus of this wasn’t to shove a “all-in-one, one-size-fits-all, plug-and-play” solution down ones throat. I know that Juniper isn’t the company to act that way and it that sense I was pleasantly surprised.
On the other hand the whole Self-Driving Network concept is still a bit vague to me. It seems (also based on their website and podcasts like this one and this one) that the whole SDN concept is a combination of several existing tools, methods and software thrown together with some machine learning in the mix. (RFC1925 rule 5 anyone?) It is promising, but we still have a long way to go in terms of standardization and processes (ITIL!) to reach that point. DevNetops is all nice, but at the end of the day changes are still cumbersome in most enterprises due to strict change control processes. At least, in my experience with infrastructure. If you fail, you fail hard and with a large blast-radius.. But I digress..
While preparing a presentation on SDN for my co-workers I am contemplating the current state of SDN in the market. The original premise of the presentations I did in the past is that SDN originated from a need to increase the agility of providing network services using a centralized model and utilizing a programmatic approach to the problem. Solutions like Cisco ACI and VMWare NSX are the prime solutions one comes to mind when talking about SDN, but when looking at the market right now, I see the following trends:
- Application modelling (one of the original premises of the Cisco ACI solution) is still a few (or many..)bridges too far for most enterprises. Most of them are still stuck with a network centric approach. Nevertheless beneficial as it increases control and speed of deployment, but it’s not using the full capabilities the product has to offer. I leave the modelling an IT landscape of over 1000 applications into security groups or EPG’s as an exercise to the reader. Every CISO wants microsegmentation, up until the point the cost become clear..
- Generic orchestration is still a big unknown. There are products out there, but it’s not clear if they are able to orchestrate a multi-vendor platform. This is something I’d like to dive into more.
- Adoption of public cloud seems to negate the need for on-site solutions. If the footprint of your local DC decreases, why invest in a complex SDN solution? Enter the whitebox. Cheap, simple and modulair. And with the adoption of AWS or Azure SDN “ships in the night” so to say. You already use it, but its abstracted away and somebody else’s problem.