Software-defined networking has come a long way since the idea was floated at the turn of the decade.
From initial exploration of the technology right through to real-world implementation, SDN has successfully made the jump from theory to practice. And with 5G, IoT and other applications now increasingly in demand further down the chain, network function virtualisation (NFV) is also seeing adoption across the ecosystem – proving to be much more than a bubble.
But what about the future? Roy Chua has been at the heart of much of this SDN and NFV journey. As a co-founder of SDxCentral, and now heading his own research and analysis firm, AvidThink, he has advised on, reported and driven much of SDN and NFV’s progress. In an exclusive webinar with the SDN NFV World Congress, Roy shared his forecasts for what the next three years will hold for the technology – here are eight of the highlights.
There will be slower uptake than expected for programmable fabrics. “There are next-generation chipsets and fabrics from the likes of Barefoot (which Intel is acquiring), Innovium, Nephos (spin out from Mediatek), and previously Xpliant (acquired by Cavium and then Marvell) that provide programmability, but from talking to data centre operators, not a lot of people know how to program them,” said Roy in the webinar. “The functions may be there, but there are limited people that can take advantage of them.“
Intent-based, AI closed loop will come – eventually. “This will eventually come - everyone’s working hard on it. There is value in it, but be careful with AI – what data sets are you using to train it, and what is the goal of this closed-loop system?”
Security and visibility will be key drivers. “Fundamentally, it’s about getting application data from one place to another in the most efficient and secure way possible to provide visibility and control – and doing this at scale.”
Going cloud native will take time. “There’s another re-architecture that will be in the works for this. To get to a cloud-native implementation, the network functions must first be broken down into collaborating microservices. If you’re a Layer 4 to 7 vendor trying to break these functions down across different product lines, maybe there should be a re-architecture - say you have a micro-service for packet classification, you don’t need a separate version for firewalls and load balancing – you might only need one. “
Kubernetes may not be the answer for CNFs. “We use Kubernetes as the answer to everything, and it may or may not be the answer for CNFs. Kubernetes was built to manage, orchestrate and schedule microservices, but for an IO-centric workload I think that augmentation and modification is needed to the scheduling or more complex placing and routing.”
SmartNICs will finally be accepted. “SmartNICs are definitely part of the equation now. The carriers I speak to are no longer resistant to them – they view SmartNICs as potentially being part of the architecture, although there are different vendors and approaches out there.”
Edge will be a challenge. “Edge computing will be a harder problem than we think. We talk about edge as the extension of all the efforts that we have in NFV and even SD-WAN, but I think we haven’t figured out all the workloads in enough detail yet. We’ll see how this evolves.”
Watch the cloud providers carefully! “Many of the changes we are seeing in SDN and NFV have come from the carriers, but a lot have come from cloud providers as well. For example, it was hard to get the carriers to take on SmartNICs until we saw the cloud providers actually using them. Telco needs are slightly different – they were built from the days of wanting five nines reliability. With these needs relaxing slightly telcos could look more like cloud providers, but I’ll be watching very carefully to see where the next generation of changes will come from for SDN and NFV.”
Click here to find out more about our 2020 event, Layer123 World Congress 2020: Service Evolution Beyond SDN and NFV.
