Evaluate Weigh the pros and cons of technologies, products and projects you are considering.

SDN, machine learning could lead to intelligent networks

The next step for SDN is to integrate network analytics and machine learning to result in automated, intelligent networks. But first, humans need to trust the technology.

The notion of open networking continues to gain traction, as service providers and, increasingly, enterprises become more comfortable with the prospect of disaggregation and commoditization.

Now, as machine learning and AI begin to evolve, the next step is for organizations to integrate networking with tools that will further automate the network, said Mathieu Lemay, founder and CEO of Inocybe Technologies, an open networking technology provider based in Montreal.

To find out more about what this shift may mean for enterprises and service providers, Lemay and John Zannos, Inocybe's chief revenue officer, discussed the evolution of open networking, the necessity to trust machines and the progression toward automated intelligent networks.

Editor's note: This interview has been lightly edited for length and clarity.

The industry has varying definitions of 'software-defined networking.' How do you define it?

Mathieu Lemay: Basically, I'll lump it all under flexible and programmable networks. For Inocybe, it's more about open networking than it is about SDN, per se, even though we come from an SDN background. It's about the programmability of the network and the dynamicity of the fabric.

Mathieu Lemay, founder and CEO of Inocybe TechnologiesMathieu Lemay

In the past, people were operating networks. Now, it needs to be machines. As we get more connected devices, we'll need to have more advanced and intelligent networks. That will have a more machine-to-machine approach to it. So, for us, SDN actually takes the underlying flexible programmable network approach. I'll make it all-encompassing by saying it starts with programmable data planes and finishes with intelligence and AI.

John Zannos: You have to know how the network behaves to have the ability to manage it in an automated way. That ultimately leads to an automated, intelligent network, where you layer in machine learning that consumes collected analytics and then directs the controller.

How far in the future are intelligent networks with well-integrated machine learning?

Lemay: The challenge with machine learning in networks is most of the network challenges we face today -- network outages from human error -- can't be learned by the machine. We have a little bit of a Catch-22. In order to have proper intelligent networks, we need to stop touching them. We need to stop -- or minimize -- human intervention in networking before we can start making the network intelligent. But the problem with operators is they don't trust the intelligence. So, there's this little circle where the network admins don't trust the software, and the software doesn't trust the network admins.

Is there a clear next step to overcome that disconnect?

Lemay: I think it will be gradual adoption. The push for cloud and IoT and the need to no longer do things manually will be a driving change, because without that [large] scale, people will always resort to manual intervention, which will continue to be an issue. I think this change in the way we live and how networking is becoming a fundamental part of our communication patterns and devices is going to drive automation. And that automation, of course, will no longer need to involve humans. That's when things will start to run smoother.

Zannos: I think we're going to see inequality in terms of adoption. We'll see some entities, companies and market segments moving quickly, just because the amount of traffic and devices will be overwhelming for them to manage manually. And I think carriers are going to move quickly, given the size and complexity of their organizations.

What are the required skill sets for this type of architecture?

Lemay: The big challenge is skill sets are changing from network admins or network operators to more of a programmer profile. But the problem is you need to find a programmer with the network knowledge or a network admin with the programming knowledge. They are two distinct sets of skills, and finding a marriage of both is extremely challenging in the industry. That has created some of the adoption barriers to these technologies.

Zannos: One of the challenges we've encountered is the networking industry has always been one of professional emphasis on vendor products. In a world where you disaggregate the appliance and bring software-defined skill sets into the equation, not only do you need a programmer mindset, but you need the individual to understand more than just a particular vendor's portfolio of platforms.

What kind of customers are coming to you?

John Zannos, chief revenue officer at Inocybe TechnologiesJohn Zannos

Zannos: First and foremost, the biggest tier-one service providers come to us for targeted help with certain projects. That's not our primary focus, though. We're seeing tier-two and tier-three carriers that are interested in our ability to simplify the problem and talk about use-case problems.

On the enterprise side, we're seeing key segments in financial services and retail. Another group we're starting to see materialize is natural-resource companies, because they're deploying localized networks. One last group, which is a specialty niche for us, is satellite and space companies that are trying to figure out a better way to manage the connection between land and satellite.

How do you work with customers' existing or legacy networks?

Zannos: At the core, we recognize the world is going to be a complex hybrid world, so we help people deploy open source [options] in specific projects. With legacy environments, there are parts customers leave alone, parts they migrate to white box switches and parts where they come to us for specific use cases. Many tier-two and tier-three carriers are more interested in starting at specific beachheads.

Lemay: You might start with more of a legacy approach to the problem, but then you'll start introducing white box switches, and then integrate with the controller and other components. That's how people are ramping up their networking; it's not a rip-and-replace approach.

What we bring is sanity and consistency in how you create all these components. You might be using OpenDaylight to build an SDN controller, an IoT controller and a LAN controller -- you'll have a variety of components in your environment. With us, customers have a consistent approach to that, so they don't end up having a variety of versions unmaintainable across the environment. The whole lifecycle is managed, as is the whole update cycle.

What else should readers know?

Zannos: I think the most important thing is the reality of 5G, IoT and cloud makes it impossible to think about a network managed manually in the near future. So, if you accept that as reality, you need to be thinking about how to get a programmable network, how to incorporate SDN, and how to ultimately get to data analytics and intelligent networks. There's not going to be a reduction in the number of network devices or in the amount of data flying across the network.

Lemay: People should start experimenting and taking bite-size approaches to the problem. I highly recommend against overall architectures that boil the ocean and solve world hunger. Instead, focus on the different use cases you're trying to solve.

This was last published in June 2018

Dig Deeper on Network automation and intent-based networking

Join the conversation

1 comment

Send me notifications when other members comment.

Please create a username to comment.

How necessary are network analytics and machine learning to software-defined networking?
Cancel

-ADS BY GOOGLE

SearchSDN

SearchEnterpriseWAN

SearchUnifiedCommunications

SearchMobileComputing

SearchDataCenter

SearchITChannel

Close