This article was sponsored by Dell EMC and written by Linux.com.
Ahead of the much anticipated 2018 Open Networking Summit, we spoke to Jeff Baher, director, Dell EMC Networking and Service Provider Solutions, about what lies ahead for open networking in the data center and beyond.
“For all that time that the client server world was gaining steam in decoupling hardware and software, networking was always in its own almost mainframe-like world, where the hardware and software were inextricably tied,” Baher explained. “Fast forward to today and there exists a critical need to usher networking into the modern world, like its server brethren, where independent decisions are made around hardware and software functions and services modules are assembled and invoked.”
Indeed, the decoupling is well on its way as is the expected rise of independent open network software vendors, such as Cumulus, Big Switch, IP Infusion and Pluribus, as well as Dell EMC’s OS10 Open Edition that are shaping a rapidly evolving ecosystem. Baher describes the progress in the industry thus far as Open Networking ‘1.0’, proving out the model successfully of decoupling networking hardware and software. And with this, the industry is forging ahead taking open networking to the next level.
Here are the insights Baher shared with us about where open networking is headed.
Linux.com: You refer to an industry shift around open networking, tell us about the shift that Dell EMC is talking about at ONS this year.
Jeff Baher:Well, to date we and our partners have been working hard to prove out the viability of the basic premise of open networking, disaggregating or decoupling networking hardware and software to drive an increase in customer choice and capability. This first phase, or as we say Open Networking 1.0, is four years in the making, and I would say it has been a resounding success as evidenced by some of the pioneering Tier 1 service provider deployments we’ve enabled. There is a clear-cut market fit here as we’ve witnessed both significant innovation and investment. And the industry is not standing still as it moves quickly to its 2.0 version. In this next version, the focus is shifting from decoupling the basic elements of hardware and software, to a focus on disaggregating the software stack itself.
Disaggregating the software stack involves exposing both the silicon and system software for adaption and abstraction This level of disaggregation also assumes a decoupling of the network application (i.e., routing or switching) from the platform operating system (the software that makes lights blink and fans spin). In this manner, with all the software functional elements exposed and disaggregated, independent software decisions can be made and development communities can form around flexible software composition, assembly and delivery models.
Linux.com: Why do people want this level of disaggregation?
Baher: Ultimately, it’s about more control, choice and velocity. With traditional networking systems, there’s typically a lot of code that isn’t necessarily always used. By moving to this new model predicated on disaggregated software elements, users can scale back that unused code and run a highly optimized network operating system (NOS) and applications allowing them to get peak performance, with increased security. And this can all be done independent of the underlying silicon, allowing user to be able to make independent decisions around silicon technology and software adaptation.
All of this, of course, is geared for a fairly savvy network department with most likely a large-scale operation to contend with. For the vast majority of IT shops, they won’t want to “crack the hood” of the network stack and disaggregate pieces. Instead, they will look for pre-packaged offerings derived from these larger “early adopter” experiences. For the larger early adopters, however, there can be virtually an immediate payback by customizing the networking stack, making any operational or technical hurdles well worth it. These early adopters typically already live in a disaggregated world and hence will feel comfortable mixing and matching hardware, OS layers, and protocols to optimize their network infrastructure. A Tier 1 service provider deployment analysis by ACG Research estimates the realized gains with a disaggregated approach to be 47% lower for TCO, three time the service agility for new services at less than a third of the cost to enable them.
And it is worth noting the prominent role that open source technologies play in disaggregating the networking software stack. In fact, many would contend that open source technologies are foundational and critical to how this happens. This adds in a community aspect to innovation, arguably accelerating its pace along the way. Which brings us back full circle to why people want this level of disaggregation – to have more control over how networking software is architected and written, and how networks operate.
Linux.com: How does the disaggregation of the networking stack help fuel innovation in other areas, for example edge computing and IoT?
Baher:Edge computing is interesting as it really is the confluence of compute and networking. For some, it may look like a distributed data center, a few large hyperscale data centers with spokes out to the edge for IoT, 5G and other services. Each edge element is different in capability, form factor, software footprint and operating models. And when viewed through a compute lens, it will be assumed to be inherently a disaggregated, distributed element (with compute, networking and storage capabilities). In other words, hardware elements that are open, standards-based and without any software dependencies. And software for the IoT, 5G and enterprise edge that is also open and disaggregated such that it can be right-sized and optimized for that specific edge task. So if anything, I would say a disaggregated “composite” networking stack is a critical first step for enabling the next-generation edge.
We’re seeing this with mobile operators as they look to NFV solutions for 5G and IoT edge. We’re also seeing this at the enterprise edge, in particular with universal CPE (uCPE) solutions. Unlike previous generations where the enterprise edge meant a proprietary piece of hardware and monolithic software, it is now rapidly transforming into a compute-oriented open model where select networking functions are selected as needed. All of this is made possible by disaggregating the networking functions and applications from the underlying operating system. A ‘not so big a deal’ thing if from a server-minded vantage point, monumental if you come from “networking land”. Exciting times once again in the world of open networking!
Sign up to get the latest updates on ONS NA 2018!