Barry McGinley: Hello and welcome to the Critical Lowdown. My name is Barry McGinley, and I'm the head of technical for EMEA for EPS Global. Today we're going to have a chat with one of our hardware vendors in open networking and some of the software within that. I'm joined today by Paul Lysander from Celestica. Paul is on the product management team in the hardware platform solutions group at Celestica. His primary responsibility is the SONiC platform. For anyone who doesn't know the SONiC platform, SONiC was Microsoft Azure's operating system which they open-sourced, and SONiC stands for Software for Open Networking in the Cloud. Paul, it's great to have you on the Critical Lowdown, and thanks for joining us.

Paul Lysander: Thanks, Barry. I'm glad to be here with you to discuss SONiC.

Barry McGinley: Great. Let's jump in. I know Celestica is strong in open networking across a range of hardware; you have one to 800G switches. From Celestica's viewpoint, how is this open approach, and specifically SONiC, enabling you to deliver modern networks for data centers and enterprise deployments?

Paul Lysander: It's interesting. We've seen a definite trend with customers we've spoken to around open networking and disaggregation. Traditionally, the networking space has been focused and driven by traditional OEMs that own the hardware. They develop the hardware and the software, and both the hardware and software are closed. But it seems that there is a trend towards moving away from that model. Certainly, the traditional OEMs still have a strong value proposition for what they offer to their customers and potentially some of their new offerings they're bringing to market. But there is a significant conversation happening in the industry from customers that are saying, "We want more choice."

SONiC has this tagline that it's the "Linux of networking." I think that's a pretty good tagline because Linux had a very similar trajectory where it brought the open operating system to customers as a choice. It's a very similar trajectory for SONiC. We're seeing a lot of customers that have been looking at SONiC as a viable choice for their own environments. In some cases, they're going to be moving full force into using SONiC. In other cases, they're going to be evaluating and potentially looking at bringing SONiC into at least part of their network.

Barry McGinley: Going a little bit deeper into SONiC, the ecosystem involves hardware vendors, software developers, and the open community as well because there's the open-source version. How crucial is that diverse ecosystem for the future, for its adoption, and how does that benefit the end user?

Paul Lysander: The openness is the strength of SONiC. Microsoft made a critical decision when they decided to open-source SONiC and hand it over to the Linux Foundation to manage as an open-source project. The importance there is, first of all, customers have a voice in what goes on in the development effort. SONiC has a very vibrant community of developers that come from a broad set of vendors. There are also customers that have a strong voice in what's happening around the entire development trajectory, so that's pretty crucial. The whole diverse ecosystem is crucial to SONiC and its adoption and even its growth.

If you look at the critical features that have come along in the last little while, it's been primarily driven around the data center and hyperscalers—you could think of them as very large data center deployments because that's what they are. It's slowly moving into the enterprise as well, and we can certainly talk about that at some point.

Barry McGinley: Yeah, because we did see that at the start. I've been hearing about SONiC for, is it a decade now? With each deployment—I know there's Alibaba, there was eBay, LinkedIn—with each deployment, they were adding more features to get it to the point where somebody that didn't have maybe 200 engineers could use it. I know we're at that point now. Do you see open networking following—you did mention it—the Linux-like adoption curve, or are there limits unique to open networking and the operations of a data center deployment?

Paul Lysander: The underlying OS for SONiC is Linux; it's based on Debian Linux. Linux has been the champion when it comes to open-source software in general. That's what kicked off open-source to a very large extent. It wasn't solely Linux, but it certainly was a big player.

The Linux Foundation, which is now the organization that manages the SONiC project, it's no accident that with Microsoft taking on the initial development of SONiC and then handing it off to the Linux Foundation, they do follow very similar trajectories. The companies that are looking at it, to a very large extent, are also, in large part, Linux users. It's very seldom that you're going to see an organization that is completely looking at closed-source in every other part of their environment but looking at open-source in terms of networking. That's probably not going to happen. In general, it's going to be customers that are open to open-source software in other parts of their environment as well. They do track pretty closely, whether it be in a broader macro scale or within an individual company or organization.

Barry McGinley: Yes, I think it did. SONiC spent a little time with the OCP, with the Open Compute Project, before it went to the Linux Foundation. And I love the OCP, but the Linux Foundation, this is what it's about. It's about open-source projects, and they run it quite well, whereas sometimes the OCP is focused on a lot of liquid cooling stuff and maybe more about the building rather than something like SONiC. I think it's in the right place now. Are there any external trends—and by that, I mean AI/ML, supply chain, maybe sustainability goals—that are accelerating interest in disaggregated networks?

Paul Lysander: Definitely. The big one is artificial intelligence. AI is front and center when it comes to what's happening in the industry overall, and all things are pointing to that trend. When you look at networking, when you think about the impact on the network with respect to data flows and the sheer amount of data that's having to traverse the network with respect to the edge coming into where the AI is being processed, it's huge data flows. There's a lot of development going on around QoS mechanisms, and even from a vendor perspective, to be able to bring higher-performance switching to the market—switch ASICs—and that bubbles up into the switches themselves. Clearly, that's a big trend.

The other one is enterprise networking. That's the other big piece of it. With the PINS working group bringing enterprise features into SONiC, which was traditionally focused on the data center, that's another big trend. I think those two trends together are bringing a lot of focus, momentum, and interest with respect to SONiC for enterprise customers. When it comes to the data center, the two sets of requirements—data center versus enterprise—are similar but different. IP networking in the data center versus IP networking in the enterprise are similar but different. In the data center, it's all about low latency, east-west traffic, and east-west security, whereas in the enterprise, the security is different. It's about user-level security, so .1x, as an example, or AAA, TACACS, RADIUS—that's not as big an issue in the data center as it would be in the enterprise.

There's a whole different set of features required in enterprise networking that are unique to enterprise networking, enterprise campus, versus the data center. There are several things at play, but I think together, they are tracking very closely with respect to efforts that are happening within working groups in the SONiC community.

Barry McGinley: And I saw the announcement by Broadcom last week: 1.6 terabit, the Tomahawk 6. I see how things are moving there. But within SONiC itself, on the feature side, what lends itself to the AI application?

Paul Lysander: That's an important question. Certainly, the QoS features like RoCE v2, which has been around for a short time now, but anything to do with how to deal with the sheer flow of traffic for AI workloads, that's a big piece of it. You mentioned Broadcom announcing the Tomahawk 6, so we'll see other vendors also bringing newer switch ASICs to the market that are higher performance and so on.

When you look at what's happening around other features like MC-LAG, which has been there for a while, EVPN multi-homing, and then there's ESI, there's an evolution happening around how to deploy ToR, leaf, spine, a three-tier or a Clos-based design into the data center. There are other sorts of features making their way in. There's also a whole effort around—because this is one of the brilliant architectural decisions that was made early on—SONiC as a set of features, at the SONiC layer, has independence from the ASIC via this layer called SAI, the Switch Abstraction Interface. That was a brilliant move in terms of the architectural decision because that brings much more choice around silicon vendors, the switch ASIC vendors. That's another piece of it. As they bring in other features at the switch ASIC level, that will start to bubble up into what's going to happen within the SONiC layer, which sits two layers above it.

Barry McGinley: It just seems crazy. We're talking about the 1.6 terabit. I think you can get 1,024 100-gig ports on it. It's just mind-blowing. Looking at a migration path for a customer, imagine we have a Juniper customer. What does that move look like for them to move to a disaggregated model?

Paul Lysander: For an enterprise customer that has a brownfield opportunity—in other words, they have something in place already—typically, they're not going to just switch over completely to not just a new vendor but a new operating system. Typically, most enterprises will start by evaluating it in a lab environment. Then they'll start to deploy it within their enterprise in non-mission-critical areas of the network. As they get more accustomed to the features, the experience, how to manage it—more of the Day 2 operational piece of it, that's a big piece for a customer. As they start to understand the tooling, what sorts of management tools they can leverage, even within the open-source community like Ansible or other tools like that, then they'll start to slowly migrate the rest of their network over to SONiC. That's the typical movement from a customer perspective. A customer that's building out their network, maybe new buildings, new locations, may choose as a greenfield location to put in SONiC from Day 0. That is also another path they may choose to follow. I would say it'll be a slow migration, but it'll be a slow and steady migration, and that's what we're seeing in our customer base.

Barry McGinley: You did touch on the enterprise stuff, so I want to get into that a little bit. You guys are strong in the data center, but SONiC is making inroads into the enterprise market, and this has been a pain point for us for a decade. We started—it was nearly all data center, and then we started to have 1-gig PoE switches, and Cumulus was starting to have a bit of traction there. Cumulus, which was eventually bought by Mellanox, bought by NVIDIA, was starting because customers were saying to them, "We want to manage our enterprise network with the same tools that we're using for our data center," because they were using Cumulus within the data center. But then Cumulus was acquired. The price point has to be good for enterprise, and the 1-gig switches were predominantly for top-of-rack in the data center; they weren't being built with PoE and what you mentioned earlier on. You've recently introduced new enterprise switches that support SONiC. Can you tell us a little bit more about your ES range?

Paul Lysander: Sure. As you mentioned, we have switches for the data center, and that movement is going well in terms of enterprise data center deployments. But we also have a line of enterprise switches that's focused on the enterprise and campus. The feature sets there are certainly going to be similar, as we talked about earlier. There is definitely an overlap between data center IP networking and enterprise IP networking, but there are features unique to the enterprise. PoE is certainly one of them. 802.1X is another. There's AAA and all of the user-based authentication features required for onboarding an end user and an endpoint in a campus environment to get them onto the network and authenticated properly. There are other things, like voice traffic. Voice traffic has to be treated differently, needs to be on a separate VLAN, so we support voice VLAN features to be able to segment and treat voice traffic differently because its characteristics are different. LLDP, the Link Layer Discovery Protocol, to be able to discover devices as they get plugged into a switch, whether it be an IP phone or any other type of device. Those sorts of things are critical for an enterprise to think about in terms of the features that are of most interest.

We mentioned earlier the PINS working group, and there are subgroups working on specific features. Celestica is certainly thrilled to be part of that initiative developing features for the open-source community. Until those features are fully vetted, because they have to be accepted by the member organizations that are part of PINS to be incorporated into SONiC's open-source GitHub repo, we still want to bring those features to customers in the interim. Our strategy is to embrace what's available in the community. If features are not available yet, we'll develop it in-house and offer it within our SONiC, and then once it's available in the open-source community, we'll make that decision to pivot over to that version of that feature.

Barry McGinley: Great. So there's a working solution as it stands now for an enterprise deployment.

Paul Lysander: That's right.

Barry McGinley: We've seen with SONiC, we've talked about AI, multi-cloud, retail, Telco deployments. We're actually running a high-frequency trading POC at the minute. What do you think is the best use case to showcase SONiC's ability and why?

Paul Lysander: Because its roots are in the data center and hyperscalers, that seems to be the one that probably resonates with most customers for a number of reasons. First, the feature set available today is very rich in the data center/hyperscaler deployments—the typical three-tier Clos design with top-of-rack, a leaf, and spine. That's probably where customers, because they do know that hyperscalers use SONiC within their own cloud environment, or a flavor of SONiC, that's probably what's going to resonate the most. The enterprise is going to be, I think, the next phase, because the features are different and they're being developed, and we do offer those sets of features within our own SONiC. That is probably going to be the next phase. I would say it's going to start in the core, within the data center, and move its way towards the edge.

Barry McGinley: Okay. That is what we're seeing ourselves as well. Obviously, the hardware and the software are separate, so they're distinctly separate, which adds, I suppose, a bit of hassle for customers. How do you guys approach this? And us being a distribution partner, is there a role for us to play in scenarios here with hardware and software? Customers do tend to say they love it, they love the separation, but when it actually comes to it, they don't want to be messing around with putting software onto boxes. I know there's no licensing with SONiC, but what is the best way of making this easier for customers?

Paul Lysander: We are happy to partner with folks like EPS Global. We see you as a value partner, and you provide a huge value add. Certainly from a fulfillment perspective, the ability to provide the end product to a customer, to partner with them around things like you mentioned—you're working on a POC, I'm sure you're working on multiple POCs. That level of engagement is critical because customers want someone that they can work with closely in those sorts of scenarios. When it comes to fulfillment, going one step back with respect to that initial engagement, understanding the features, understanding it to the point of actually testing it, kicking the tires on it, doing a POC, and then actually deploying it—that's where partners like EPS Global are so critical to making a SONiC deployment successful.

Barry McGinley: Okay, great. Looking at the future a little bit, I see you guys won a Dell'Oro award for your 800-gig switch. We've looked at 1.6-terabit. I don't think you're going to be making an announcement today, but what's in the pipeline? What's coming down the road with Celestica hardware and SONiC?

Paul Lysander: The DS5000 is our flagship high-end switch today, which is based on Tomahawk 5. As our silicon vendors that we work with release the next-generation ASICs, we'll certainly work on bringing products to market that leverage the latest and greatest technology. That's on the hardware layer. From a SONiC perspective, we will continue to work within the community to bring the next-generation set of features within SONiC itself, both for the data center and the enterprise. Those are both parallel tracks for us where we're certainly working within the community to leverage what's happening with community members around features for the data center, and for the enterprise, as I mentioned, there's a whole separate track around bringing the unique set of features required for the enterprise campus environment.

Barry McGinley: Which area are you most excited about going forward? Is it the enterprise? You seem sweet on the enterprise play, so is that where you see the most gains coming down the road?

Paul Lysander: I'm excited for both tracks, honestly. SONiC has already been in the data center and hyperscaler environment, so that's already well-known. I think the enterprise play is where it's less known, and that's where there is going to be greater momentum within the community from the PINS working group. It's a matter of exactly how that's going to play out. We're certainly focused on both—the data center play as well as the enterprise, which is why we have a distinct set of products that play into both those environments.

Barry McGinley: I agree. It's another string to our bow. We haven't had an enterprise play for a long time, as I said, since probably seven or eight years since Cumulus, so it's great to have it. Okay, so finally, for our listeners who are maybe considering Celestica with SONiC sitting on top, what advice would you give them for a successful engagement?

Paul Lysander: The first thing is to get educated on what SONiC provides. That's the first step. There are lots of resources. There's even, within the OCP you mentioned, the OCP Global Summit that's happening later this year in San Jose has an entire track just on SONiC. For past events, those recordings are available on YouTube. The first step is to really get educated on SONiC and what it provides. If someone who's interested wanted to research it, they would see that there's a huge and very active community, both from a vendor perspective and a customer perspective, that is eager to deploy SONiC within their own environments. That's the first thing: the education piece.

The second piece is to actually try it, try it within the lab environment, to work with folks like yourselves to do POCs, to replicate their own non-SONiC environments within a SONiC POC to see how those features work, how SONiC behaves, and what kind of tools are available to manage SONiC. We mentioned Ansible earlier. There are other open-source solutions as well as partners that offer orchestration solutions for SONiC. There are also partners that provide their own software solutions for easing the migration from a proprietary NOS over to SONiC. They actually help to do that migration for configurations from one OEM vendor over to SONiC. There are a lot of things in play there. I think education first, kicking the tires, trying it out in the lab environment, maybe within part of the network, they can start to experience how SONiC behaves and works, get a very solid comfort level, and then start to do a broader deployment within their enterprise.

Barry McGinley: Yeah, I'd echo those sentiments. There really is a full ecosystem there now with orchestration, good hardware, and well-featured software. Final question, I always ask this one: where do you see open networking in five years' time?

Paul Lysander: Five years is a long time. It's hard to imagine five years, but I'll just say even in the short to medium term, it's going to continue to gain momentum in the industry as well as within enterprise customers. From all that we're hearing and all that we're experiencing ourselves within our customer engagements, there is a greater interest, not just at the periphery, but in looking to do POCs and testing. There's a greater interest for SONiC this year than there was even last year. In five years, we anticipate that there'll be that much more interest and that much more traction with SONiC around these large enterprise customers as well as small-medium businesses.

Barry McGinley: Brilliant. I'm looking forward to the next five years. Paul, thank you very much for your time today. That was really enjoyable. Thank you.

Paul Lysander: Thank you very much.

Barry McGinley: And for anyone that wants to get in touch about, as Paul said, the journey for open networking, educating yourself, and once you're ready to move to the next step, EPS Global will be happy to help you move from there.