Customers are leveraging Cyxtera’s software defined network service, CXD, to effortlessly balance in a scalable hybrid environment. Holland Berry, SVP of Strategic Alliances at Cyxtera explains how Cyxtera is applying their revolutionary CXD platform to give enterprises access to a scalable, seamless hybrid environment.
INTRO: [00:00] Welcome to the Tech Deep Dive podcast, where we let our inner nerd come out and have fun getting into the weeds on all things tech. At Clarksys, we believe tech should make your life better, searching Google is a waste of time, and the right vendor is often one you haven’t heard of before.
Max: [00.18] Hi I’m Max Clark and I’m here with Holland Berry, who is the Senior Vice President of Strategic Alliances for Cyxtera. Holland, thank you for joining!
Holland: [00.25] Max, thanks for having us, good to be here!
Max: [00.28] So Holland, we were just chatting a little bit offline but you have an interesting path into Cyxtera that started with an acquisition, and landed you in the cybersecurity division with Cyxtera and now onto managed services and alliances. Can you give me a little bit of background of how you ended up where you are now and what your path has been?
Holland: [00.51] Yeah, absolutely. So, I’ve been in technology for all of my adult life. I started in my teenage years, actually, I started a little IT consulting company with my little brother, and twenty-seven years later – gosh, I’m dating myself, but here we are! So more recently I was part of a little cybersecurity setup, it was called Kapper(?) and we made a hypervisor-based security appliance, that I’ll call the cool stuff… I won’t drain our time with that, but that was one of the initial companies that was acquired as part of the formation of Cyxtera back in May of 2017, so I started out on the cyber side with the company and saw some of the cool stuff that we were working on to modernize the datacenter, make it feel more like a cloud experience, create an SDN fabric, I know we’re going to talk about that later, and so… As I was exposed more and more to what was happening with my friends over on the datacenter side of the building I thought, “Oh gosh, this would be a nice opportunity to go and help out over there.” So, here I am!
Max: [01.47] Cyxtera is a relatively new company, but it’s a collection of assets from really old companies with a new brand and its spinouts, so what is Cyxtera now and where did it come from and what are these things that pieced it together?
Holland: [02.01] Yeah, absolutely. So, we started out back in May of 2017 and you know, at our core we’re an enterprise colocation and connectivity company, and that was largely based on the acquisition of the CenturyLink Savvis datacenter footprint. So, we’ve got sixty-two hybrid ready facilities across twenty nine markets, we’ve got about three million square feet of datacenter space, and in the neighborhood of two hundred and fifty megawatts of power capacity. We’re the third largest provider of internet connections globally, it’s somewhere in the region of a hundred and forty network providers with an average of fourteen, fifteen per facility, and somewhere in the neighborhood of forty thousand total cross connects, so a pretty big focus on the interconnection piece of our story. I should also note too that the majority of our datacenters are within three milliseconds of public cloud providers, so heavily cloud adjacent and when we say hybrid ready, that’s part of that statement is the fact that we know people are going to be spreading workloads between private cloud environments, colocation environments, and that adjacency and access to public clouds is important, because we know applications or application tiers might live at both ends of that proverbial fence.
Max: [03.15] Cyxtera has a service line called CXD, and this is your… I mean, it’s a Nutanix-based platform, and this is your hybrid cloud, bare metal, you know, insert marketing term here, but how would you actually describe CXD in Cyxtera speak?
Holland: [03.33] Yeah, I’ll try to uplevel it and make it as simple as possible. So, at it’s foundation that modernization that I talked about, making our datacenters a little bit more cloud-like in terms of how you consume the services, foundationally that’s delivered by a software programmable network, an SDN fabric. That’s really the foundation of CXD, connecting one point in a datacenter to another point in a highly automated, software programmable fashion. Sitting on top of the CXD network fabric, and CXD by the way stands for Cyxtera Extensible Datacenter, is our enterprise bare metal platform. We launched with Nutanix, we’ve added some additional entities on the fabric which I’ll talk about in a second, but just think about it as an SDN fabric with infrastructure services and an ecosystem of partners that sit on top, so if you’re one of our colocation partners or customers or someone new to Cyxtera, you have this great platform that you can run your applications on effectively.
Max [04.39] So CXD and the approach between CXD and this concept of enterprise bare metal and hybrid environments… You know, the trend was very much, go to the cloud, and now companies are realizing I think that maybe going to the cloud isn’t necessarily what they want, and they want something a little bit different. So, how do you view that something different and what is Cyxtera trying to accomplish with CXD?
Holland: [05.00] It’s exactly what inspired us to create the platform and I wouldn’t necessarily describe us as an alternative to public cloud, although sometimes we are for people, but I like to think of us as a compliment, but Cyxtera is a big consumer of public cloud. We had the conversation internally when we first started, it’s like, “Hey, we’ve got a few acquired companies, with a mix of infrastructure all over the world, right, across multiple platforms, what makes sense to run where?” And we came to the conclusion that there was space in workloads that were perfectly suited to run up in public cloud and we still run them there today, even though we could easily run them all in these big banks of infrastructure and facilities all around the world, but if I need to go and do a ten thousand VM scale test for one of our DevOps groups for a new piece of software that we’re writing, that exercise is much better suited in the public cloud. But for a steady state, kind of back-end IT apps, things that are very demanding on the metal, things that might get you into some cost considerations running up in public cloud, we wanted to introduce another ways to think about it, so one of the design goals with CXD enterprise bare metal platform is to give our customers, you know, all of the financial and operational flexibility of cloud, but combine that with the control, performance and security elements of running your local dedicated on-prem or colocated infrastructure. And so we’ve kind of created this third category where it’s not something that you’re running in your campus datacenter in terms of infrastructure and it’s not something you’re running up in public cloud, it’s this thing kind of in between and that’s in its essence is what CXD and enterprise bare metal is.
Max: [06.45] So what I’ve noticed is that it’s not necessarily a financial… When you say financial or operational consideration for public cloud versus something else, in almost every scenario it seems that moving to a public cloud environment is more expensive for an enterprise to take on. Right, so the decision for most… I wouldn’t say is we’re trying to save money by moving to the cloud; we’re trying to do something else or accomplish something else, and traditionally cloud is you know, elasticity and unpredictable workloads, you say you want to scale up a ton of VMs in order to do a DevOps test and turn them off, you have that kind of aspect towards — it’s a managed platform, you know, you want a managed database or you want a managed objects store, something like that, right? But what I’ve noticed and what we’re seeing and I’m curious about if you’ve seen it as well, is it’s really more about the operational management, you know? Companies don’t want to scale operational overhead to support lots of infrastructure globally, and cloud made that really easy for them, like, “Oh, we don’t need to figure out how to put somebody in Hong Kong or fill in the blank,” right? So today when you look at CXD and enterprises coming to you, how much of it is financial now and how much of it is operational and what are people actually trying to solve that they talk to you about?
Holland: [07.57] I think it’s a combination of things, I think cost savings is the topic dejeure, just because a lot of people had and still do have a cloud-first strategy, and what they interpreted that to mean was, go to public cloud. We have a slightly different take which hopefully we’ll get to in a second, but they got everything up in public cloud, they did our cost analysis and all of a sudden the CFO arrives with some questions, because the cost predictions and then the actual cost didn’t line up. And so now we’re seeing a trend of repatriation that is squarely on cost, but you’re right – there are some wonderful attributes of public cloud, especially the platform elements; you get a whole suit of capabilities around just, you know — we’ll take Amazon as an example, the EC2 instance, right? You’ve got hundreds of services that surround that instance to make it really easy to do kind of whatever you want, and I think that element — that’s one of the reasons that it’s attractive to us too and why we run certain workloads there, as well. But, there are elements of — where we’ve talked about them before — performance, there’s some elements of security, and we still that being cited as a reason why people continue to run on dedicated or colocated infrastructure, so I wouldn’t say it’s any one thing but I do think cost is driving a lot of the conversations, at least the ones we’re participating in.
Max: [09.20] When you talk about cost driving the conversation, it’s not necessarily the cost of the VM, it’s all these ancillary costs that people aren’t factoring in that come a long way with the VMs, it’s like bandwidth, you know?
Holland: [09.30] You’ve got that right, yeah; or maybe they selected the wrong disk type to pair it with or wrong storage volume, maybe they’re blowing through their CPU credits like they didn’t anticipate. Maybe there are additional whatever that service is that was necessary that they didn’t anticipate out of the gate. There’s all these service adjacencies around the actual VM itself that can run up the bill pretty quickly, and when you’re running a private cloud on dedicated infrastructure, your costs are a bit easier to predict, we would argue.
Max: [10.01] Right, because you know how many nodes you have, what those nodes cost, if you’re over subscribing or under-subscribing those nodes, what your — I mean, it’s not variable at the same degree, right?
Holland: [10.11] Yeah, that’s right – and for us and the way that we offer the service, we’re not metering anything, right? So you can pin a bare metal compute node app redline, a hundred percent CPU, a hundred percent memory, 24/7 and you know, that’s on you. It’s the same thing with our network bandwidth, and those are things that can get you into trouble if you didn’t predict that steady state of a workload or bandwidth utilization in a public cloud environment. So, we’ve been particular in the way that we deliver our platform, it’s highly predictable from a cost perspective.
Max: [10.40] Can you walk me through what the CXD platform components are now, and what the selection somebody would go through in deciding, you know, “I want option A or option B or option C?”
Holland: [10.51] Yeah, absolutely. So, as you mentioned at the top, we launched with Nutanix, we have a heavy slant on our catalog with HCI capable configurations to run the Nutanix software. We broadened things out a little bit in the recent quarters, so we have now options available from Hewlett Packard enterprise… Most of those configurations in our node catalog are also HCI capable, you can just as easily run the Nutanix stack on top of them, you could also run VMware’s VSAN if you wanted to. We have options from Fujitsu as well, in our EMEA market, specifically in our London market right now, and those as well have an HCI capable configuration to run the Nutanix stack as well. We have some upcoming things that we’re going to be announcing with Nvidia by the time this publishes, so I think I’m comfortable talking about it, we’re doing some really interesting things about GPU as a service and Nvidia’s DGX platform to further thicken up the capabilities on the platform. But the core concepts all remain true, right? We want to give folks an easy way to consume enterprise-grade infrastructure, allow you to consume it and pay for it just like a public cloud instance, and give you a dedicated single tenant access to it, and when I say dedicated I mean all the way down to the BIOS or the IPMI. You could make system level changes that could technically brick the box, but we want to give you the experience just as if you had bought it, and racked and stacked it in a cage that you own or operate. And so, that’s the high line of some of the technology options that we’re delivering, that are Cyxtera branded. And then to compliment that, we’ve partnered with a number of companies and that community continues to grow that also has services that sit on top of the CXD networking fabric. So we’ve got a great storage partner called Zadara, if you need to go get a petabyte of object storage for a week, they can do that, and you can plumb the network so… Maybe you’ve got five HP ProLiant DL360 servers from Cyxtera, you can them pair that storage from Zadara, all using that same automated networking plumbing, and make that environment look like one logical together stack of infrastructure, just as if you went and built it yourself, but what we’ve saved you from doing is actually having to not only define and buy cage space within one of our colocation facilities, but also procure that IT gear, get it shifted, rack it and stack it, you know, provision it, all that kind of stuff, and most of us know we’ll see in sixty to ninety days when you start that process. You should be able to stand up a datacenter environment within a day or two, and leverage all these on demand services that we’re leveraging as part of our platform, as part of our marketplace.
Max: [13.44] And for clarity, when you say HCI, you’re talking about hyper converged infrastructure where the node has a compute and RAM and storage, and everything is in one box, you can build blocks with those boxes together?
Holland: [13.55] You got that right, and sorry for the acronym soup there, I have a tendency to go there, so… We’ll unpack those and let them fly.
Max: [14.04] You know, you talk about datacenter, how quickly can you build a datacenter and stand up teh datacenter, and what is the actual impact to companies, right? So, datacenters are expensive to build, just the floorspace is expensive to build, and enterprises are still maintaining a computer room, I mean that’s a lot of investment in order to build those things out and of course they’re sticky and they’re fixed capacity and you can only have so much power and expansion is really hard, so that was the initial push into colocation markets, right? You know, “Hey, it’s expensive for you to build your own datacenter, so come to our datacenter,” and then that became, well… It’s expensive, or complicated or time consuming to maintain server infrastructure, so go to the cloud, but then the cloud becomes expensive, so now it’s… Okay, we’ll kind of come back and you’re one foot in a datacenter and one foot in the cloud, right? That’s basically what we’re talking about.
Holland: [14.49] Yeah! And look, we’re saying do both, just like we’re doing – we’re practising what we preach. That’s exactly right, and to the extent we can remove some of those complications of having that dedicated infrastructure while retaining some of those benefits that people have become accustomed to with public cloud, I think we’re going to delight our customers.
Max: [15.11] So I mean, operationally you get into all the fun things of, okay you have to ship how many pallets of gear to a facility in Ashburn, Virginia, and then hire somebody locally in that market who can go and unpack those boxes and cable those boxes, and whoops, you forgot to buy DAC cables, so you have to wait for those to show up and you know, or the FSE’s incorrect… You know, what I think people see a lot is just the warranty management of equipment cycling, “We’re two years into this box and somebody died on the motherboard, who’s going to be there to let a tech in to replace this motherboard component?” So, in this with CXD, you’re taking all of that off the customer’s plate, and they get resources but they don’t have to physically manage those resources.
Holland: [15.54] That’s absolutely correct, and there’s one other element that I think is important to point out when you’re talking about that DIY approach of building out a colocation cage in something like a CXD platform, which is that notion that you sort of have to build to a perceived peak on what the demands of that application are going to be, out of the gate and then maybe a few quarters into the future. And with CXD on our enterprise bare metal program, you can break the bulk of that typical build to peak IT procurement and design process and just order a node at a time, so maybe you’re standing up a DDI environment and you know out of the gate there’s going to be five hundred users, well you might only need eight of our servers to do that out of the gate… But you know, you’re in growth mode, you’re hiring a bunch of folks and you know that environment and that footprint is going to need to expand, up to maybe the equivalent of needing twenty servers over the course of the year. Well, most people are just going to buy those twenty servers, rack them in a cage and have some of that infrastructure sitting idle and have that capital sitting idle as well when it could have been deployed elsewhere, and our thesis is that if you utilize the way that we’ve built the platform, start with that minimal set of servers and if you need one tomorrow, cool, go to the portal, click a few buttons and you can get one. Then if a month later you need three more, the same exercise; that capital that you would have deployed at that build a peak exercise up front can hopefully be used for more interesting things.
Max: [17.18] So what are the contract cycles with that? So if you’re going in and you’re saying I need to add a node, and I go and click to add a node, is that something that’s provisioned on a month to month basis, I mean, you can get into terms for discounts… How does somebody consume this?
Holland: [17.30] Yeah, the typical term for one of our enterprise bare metal nodes is a year. If you think about the profile of applications that are running on these, these are longstanding, steady-state, back-end IT, critical production infrastructure, although we do have some DevOps folks on the platform as well, and it’s perfectly suited for that… We do have some shorter terms under a year, but typically a year is the default and you know, with three years not being uncommon either. Obviously there are discounts if you commit to longer time periods.
Max: [18.02] Nutanix was a hyper converged infrastructure box, right, this HCI infrastructure, but then they also had this idea that you’ll bring your own platform on top of it, you’re going to run VM on top of Nutanix, or something else, or Kubernetes is really popular… So, when somebody’s making that decision of, “Do I want to run Nutanix or HPE or Fujitsu,” is this something that they’re selecting or are you talking to them saying, “What are you running, do you want to run Kubernetes, you should run this because it makes more sense for you.”
Holland: [18.30] Yeah, you know, just like we’re the infrastructure folks right, so that persona of saying, “Gosh, building a datacenter is hard,” that’s really our expertise, right? We’re great at power and cooling and air displacement and physical building security and all that stuff, and I think we’re pretty good at the physical infrastructure piece as well, so networking and physical hardware boxes… As we start to move up the stack though, we tend to allow our customers to make that choice. We have opinions, we’ve got great engagement from the OEMs that we work with, the HP, Nutanix, Fujitsu, et cetera, but once we start getting into the hypervisor and the workloads above, we tend to start bringing in the folks that are just simply better at it than we are. So, what we do in our portal or with our complimentary set of APIs that we use for provisioning, is we’ll allow a user to select a hypervisor; right now we offer Nutanix’s Acropolis and VMware, and we will include those images – we’ll image the boxes with those hypervisors. You can also choose a no OS option. And that’s just the kind of get the box — you know, bootstrapped and up and running, so especially if you already have an existing environment, a colocation or existing enterprise bare metal, you can quickly add that resource to the resource pool and start motioning VMs over to it. We do that more as a courtesy and less as something in an advisory role. So, we give you the choice, but it’s really on the customer and their partners to make that decision on what the best route is, and there’s merits to argue all day long about why VMware versus Acropolis and the Nutanix ecosystem, I think Nutanix has done a fabulous job in building up feature parity with what VMware has built. As you know, lots of people have written automation around the VMware APIs and there might be certain features that they just need in one platform versus the other, and we definitely leave that to the customer.
Max: [20.29] I think that also probably depends a lot on — I mean as you say, it depends on the customer, what the customer’s existing application is, and what they’re managing. I mean, you get into like web scale environments or data science, it’s not going to be running on VMware or Nutanix, they want to do something different. If you’re giving them IPMI access, you’re like, “Okay, here’s IPMI, go do whatever you want with this box now,” that’s what you’re saying?
Holland: [20.49] That’s exactly right, yeah. And so for those folks that are running Kubernetes and Docker, you know, whatever container flavor they like, they’re going to be throwing Centos or Ubuntu or Route or something on the box and we don’t want to be so protreptic as to say, “You must use this environment in this way,” and that’s been our philosophy with the platform in general, right? We don’t want to force you to use our VLAN scheme, right? All of the Layer 2 networking segments that the customer gets to define are encapsulated within an VXLAN, so they can bring their own VLANs, bring their own Layer 3 space, bring your own IPs, public or private; we support BGP… So really, this is a bit of a blank infrastructure canvas for our customers to build whatever they want on, and we’re trying not to put a ton of guardrails around that, so people don’t feel constricted with how we design the platform.
Max: [21.37] Is CXD available now across Cyxtera’s entire environment, your portfolio of datacenters or are you still in the phase of rolling this out?
Holland: [21.45] We’re still in the phases of rolling it out, we have it available… I think we’re in eight markets now, and when I say a market, that market can be comprised of multi datacenters, as an example out here in silicon valley we’ve got seven datacenters, right? So, when we roll it out to a market, we roll it out to all of the datacenters, so we have dark fibre running between all of our facilities on the platform at present, as we have access to, you know, whether you’re in datacenter A or datacenter B, it’s all available: the compute infrastructure is available, the network’s available, our partners are available, so… I would say over half of our datacenters have it right now, and almost every single of one of them is on the roadmap to get it.
Max: [22.26] If I was an existing Cyxtera datacenter customer, it sounds like moving into CXD would be very easy: create a cross connect in the facility, plug into the platform, now I have additional compute nodes available to me that we can scale up as needed, or scale down, whatever our profile looked like to our existing infrastructure. If we are outside of a Cyxtera datacenter and we’re in a cloud environment or maybe we’re in a different facility, what does that interconnection or that networking side look like? Is that driven to public interconnect, is it driven to some sort of a cloud interconnect fabric, or are you talking about bringing in additional layer 2 transport circuits between buildings… What happens?
Holland: [23.03] Yeah, so the easiest way to get connected into the environment – and look, we have a multitude of ways to do it just given the footprint of the network providers that we have in every building – but the easy button, and we want to make sure we have this available out of the gate, is we’ve made available as part of a core CXD offering, a Cyxtera branded service called IP Connect. This is a blended bandwidth service that’s behind the scenes combined of two top tier providers, so think like Level 3 and CenturyLink — We can tie that, in a software defined fashion, back to the fabric and pair your dedicated infrastructure to it, and then you can use Layer 3 type services to get into those nodes. That could be landing an IPsec tunnel or setting up on VPN server, that could be using NAT and opening up SSH, so you know, Port 22 can come straight over the internet, some security folks might wince at that, but we give you those options. So, it’s the combination of the CXD networking fabric, the bare metal sitting on top, and then that IP Connect service. All those ingredients mixed together will allow for a non-Cyxtera customer to come in and get access to that newly provisioned gear and then pair it with whatever environment they want to in a multitude of ways, but that IP Connect service is the easy button.
Max: [24.24] When we go through datacenter sizing and costing exercises, you get into how much floor space do you need, how many racks do you need… There’s a lot of moving pieces, server equipment, which is — and you get into some point operational overhead, you know? What is your human cost to these things, what’s your warranty, what’s your logistics, how many spares do you have, that list just keeps going and going and going. But if you look at it and you say in a normal cycle a company is buying a lot of gear and they’re leasing it for three years but they’re assuming a four or five year lifecycle for that gear, at some point that equipment comes out of a cash flow. There’s not a cash flow component to maintaining that equipment for hte equipment anymore, there still is for the datacenter and everything else, but the financials change. They’re pretty significant when you look at a four or five year total cost for a server in a cabinet that changes. So, when we talk about CXD and we look at these things and look at your terms, how do you match those with customers with what they’re used to from a finance perspective, and as equipment ages, are you replacing the equipment, are we talking about refresh cycles, are they talking about, you know, at some point they own the gear or the cost changes… What can somebody expect?
Holland: [25.34] So the platform, I just wanted to state that it’s relatively new, I think we’re about eighteen months into having it in production and actually have a web GUI, so we haven’t had that first refresh cycle, but what we’re anticipating is around that three year mark, people are going to start saying, “Hey, I want to get the new thing,” right? So, the plan is to refresh our node fleet, have the new things, and then instead of the traditional IT procurement motion, which is, “I’m going to go buy the new ones,” we’re just going to allow people to trade them in and then motion their VMs, or they’re not using VMs, you know migrate their environment over to the new stuff, and so that saves them from year three, a spike in that capital outlay where they need to go buy the new stuff. We’re saying that we already have it there, you’re just going to subscribe to the new stuff, re-up on your term – whatever that was – and motion your environment over.
Max: [26.25] I mean, has there been discussion about maybe people want to stay on the hardware and what that would mean in terms of financial impact for them?
Holland: [26.33] Yeah, I mean staying on the old hardware that’s sitting on our enterprise bare metal –
Max: [26.38] Yeah, so like if it was your equivalent, your bare metal, and the other customer who has been on that infrastructure for three years, you know, and what happens like year four? Maybe they say, “This is working fine, can we get a better financial – the equipment has been depreciated, it’s been there for a long time…” –
Holland: [26.58] Absolutely, and since we haven’t hit that yet we haven’t gone through this, but what I anticipate that we’ll do is if someone says, “Look, I love this gear, it’s working fine, can we renegotiated, maybe get a better rate,” I think we’re going to be absolutely open to that conversation. They’ll have the choice and what we’re certainly not going to do is say, “Sorry, we’re bringing in all the new stuff and your stuff’s out, you’ve got to use the new stuff,” because we know that there might be some functionality on a certain level of hardware that might be critical to running their application and we’re not going to put them in a bad spot, so… I think you hit it on the head that we’ll be having a conversation with these folks on the last generation of gear, and it’s probably going to be financially beneficial to them if they choose to do so.
MID-ROLL: [27.39] Hi I’m Max Clark and you’re listening to the Tech Deep Dive podcast. At Clarksys we believe tech should make your life better, searching Google is a waste of time, and the right vendor is often one you haven’t heard of before. With thousands of negotiated contracts, Clarksys has helped hundreds of businesses source and implement the right tech at the right price. If you’re looking for a new vendor and want to have peace of mind knowing you’ve made the right decision, visit us at Clarksys.com to schedule an intro call.
Max: [28.05] So I mean, network takes some time to establish and figure out, you know, what kind of connectivity we have, are we bringing up a Layer 3 service or an internet product, do we need private connectivity… But I mean, how long does it actually take to provision CXD from scratch? Like, we say we’re going to go, we’re not a customer yet, we want to turn it on – what does that cycle actually look like in the real world?
Holland: [28.28] Yes, I’m going to start out and answer with an answer, it’s the start of all great answers which is, it depends! So, I would say… One of the longer pieces in that process has nothing to do with technology, it’s all about legal contracts and MSAs and lawyers sending back and forth redlines for contracts and things like that.
Max: [28.52] Well come on, lawyers never take a long time with these processes, let’s be truthful…
Holland: [28.55] I know, I know, I don’t want to insult any of our lawyer friends in the audience, but the truth is, the larger the organization the longer that’s going to take, right? The lawyers are going to find a piece of language they want to argue about with us and that’s fine, we’re used to it but it’s going to insert massive latency into that spin up process. So, assuming we can get through that MSA contract negotiation phase relatively quickly, and we can kind of get them onboard as a customer, which we can do relatively quickly, you know… We could stand up somebody in, let’s say… I’m going to say ten days, so ten business days would be a good one, and I’m factoring a few things in, right? It’s a new platform, there’s training, there’s always some discovery, I’ll say, around connecting someone’s VPN since they’re going to be coming in externally, right? We’re getting that external connectivity going, train them up… So, being up and workload ready, assuming that we can shorten that legal process time, maybe down to a handful of days, we can get your workload ready in ten days from a technology standpoint, and let me give you the alternate of that. If you’re already a Cyxtera customer and you have one of our CXD ports, which is your gateway out to the fabric, I could get you a piece of infrastructure in under an hour. I can provision an HPE node and get a Layer 3 address on the iLO port and stretch a network from your cage to your CXD port… I mean, it’s literally an hour. I can provision one of those IP Connect services to get you out to the internet, up to five gigs, in about sixty seconds. So, it’s not really a technology conversation in terms of timeframes, it’s all the other stuff.
Max: [30.34] So IP Connect is interesting because Cyxtera at inception was carrier neutral. We talk about carrier density and providers that are in your facility, but not everybody wants a carrier neutral environment, right? I just want you to take care of everything for me, so I assume IP Connect came out of that place of how do we give a customer the ability to not have to think about this stage in rapid provision, and then you said up to five gig… You know, what are the boundaries of IP Connect and at what point do we go back and say, “Okay, you need a hundred gig interconnect, go talk to CenturyLink or go talk to Azayo(?) or go talk to somebody else…
Holland: [31.09] Yeah, you’ve hit the nail on the head with kind of the easy button and with, “Hey, we don’t want to think about this piece, provision it,” but I don’t see a lot of our customers using our IP Connect circuit as like, their primary method of delivering this application. Although some might, and look, it’s redundant, it’s going out to two different providers with redundant fibre PAS and everything is redundant, even to the CXD port, it’s a dual-fibre handoff, so you have redundancy kind of all the way outside of our walls and through however the provider has architected their fibre network. That being said, we’re seeing IP Connect used more as a compliment to existing connectivity, especially for existing customers, right? So maybe they’re experiencing a service degradation with their main IP transit provider. Sixty seconds, we can get you spun up and you can have another path into your environment or another way to make your application more available. I see it more as a compliment to the buffet of carriers that we have in our datacenter and less of it as an alternative to… Although some people might be using it that way. And at the end of the day, it’s still traversing at least two of those existing carriers, so it’s not like they’re being cut out of the equation, it’s just the path in which that traffic is taking to get to those carriers might be a little different.
Max: [32.22] In the past, Nutanix wasn’t necessarily a super flexible hardware platform and it had a relatively defined spec of what a Nutanix node looked like… HPE, a little more flexible, and when you start talking about a customer application and a spec for hardware, how flexible can you be in provisioning infrastructure and then if a customer has a custom built that they want, a very specific thing… At some size, they’d say, “Hey we want a thousand physical machines,” I’m sure you would do it, but I would assume it also impacts your ability to rapidly provision additional nodes, so like… How do you balance those two and what’s available for people?
Holland: [33.01] It’s a great question, when we went to market with this to see kind of what people are using, what are maybe some common configs to support the applications running on top, and let’s build a standard node catalog that’s available on demand to match those application profiles, and you know, we started with one thing and we got smarter as we engaged more customers, and we made some augmentations to the catalog, and that included the introduction of a small handful non-HCI variants. We noticed that some people just needed a utility node, they didn’t need all the functionality or capabilities with the hyper converged stack, they don’t need a ton of local disk, they’re not going to be pairing it with anything else, this is just a low impact, low criticality server. So, we’ve introduced those options as well to the catalog, and so that’s kind of what’s standard right now. I think we have somewhere around fifteen different configs that hit the upper end of all available configs on a compute node, they’re more enterprise grade, and that’s why we call the platform – in part – enterprise bare metal. These are not low core count, low memory, inexpensive servers, I’ll say these are very much tooled for the enterprise. So, you know, that’s what we stock, that’s what’s available on demand, but we frequently have conversations with customers saying, “Hey, this doesn’t hit the mark, we need something else,” and we don’t really have a size threshold if someone wants something custom. They’re going to lose out a little bit on that on-demand portion or benefit of the platform, they’re going to lose out a little bit on if they needed to rapidly expand, say they wanted six custom nodes right, that didn’t conform with the specs of our node catalog, and they needed to add three more in three months, that’s something that we’ll go buy for them and rack on the fabric, it’ll just take thirty days to do so, they’re not going to be able to get it in an hour. That’s absolutely fine, and a lot of customers are totally comfortable with that, some customers say, “Hey, we’re going to build custom nodes and we’ll give you a forecast,” and that’s okay too… So, we certainly don’t need a big dollar figure to go have a custom node conversation with the customer.
Max: [35.06] Something I’ve seen a lot when people start moving out of a cloud environment back into a physical environment is sizing. You know, not really knowing… I mean, none of these public cloud vendors make it really easy for you to understand what the underlying infrastructure hardware is, you know, that you’re running on, right? They make it some way, ESP or some other kind of esoteric terminology, right? So when we go into these physical — back to physical migrations, you get into the dialogues of, how many CPU sockets do we need, how many cores do we need, are we running Intel, are we running AMD, how much RAM do we need per core, how much disk to we need, is it SSD, is it not, is it nVME? How much network do we need? Is it ten gigs, twenty-five, what does the actual stack look like? As somebody is going through that transition, a lot of it becomes a little bit of experimentation, like… We’re going to run this application this box, how does it work? Oh, we need to tweak these noms a little bit. How do they go through this cycle with Cyxtera?
Holland: [36.00] Yeah, so luckily there’s been a host of solutions that have popped up to help with this process, right? So you can size something for the three big public clouds, then you can also get the equivalent of what you need if you’re running on-prem, you know? We’ll forget about the hypervisor type for that part of the conversation… But in addition to that, and I’ll use Nutanix as an example, they’ve got a great sizer tool – you go to their website, plug in a bunch of stuff and they can give you a rough idea, based on a number of inputs, what type of box you should get, factoring in over subscription and all the things that you’d expect. So I think today it’s much easier than it was enough a couple of years ago when people were like, “Gosh, I don’t even know where to start and I need to run this thing,” so I would argue that there’s great tools out there to help make an informed decision.
Max: [36.51] And so in that case, if they came in and said, “Okay, we’ve got this — we want box 2C in your catalog,” and they start there, they run a workload for thirty days and they say, “This is not the right size hardware for us, we want to switch.” Is that something they can do?
Holland: [37.08] Yeah and typically in those situations when they’re not absolutely sure about what they want, we’ll say, “Hey, we’ll do a thirty day trial, we’ll get you stood up, you don’t have to pay us anything and let’s see how it goes, if we need to change, we’ll change.” So, we typically embrace that collaborative spirit on, let’s get it right out of the gate and not make it hard for you if we got it wrong for some reason.
Max: [37.29] What size companies are using CXD today, and what’s an average environment for you now? We’re talking about a handful of nodes, dozens of nodes, hundreds of nodes… What are you seeing deployed in real life?
Holland: [37.41] It’s definitely on the lower end of the scale, it’s not hundreds, although we do have some opportunities we’re working on that are very much in the hundreds, but since it’s a newer platform and since a majority of our current customers, we have about 2300 customers spread across the planet, tend to be on the larger enterprise size of things, eighty percent, I think most of our customer base is enterprise… And I think we’ve got 33% of Fortune 100 sitting with us… So if you think about their appetite for a new cutting edge technology, it’s typically a little bit later in the adoption cycle, so we get people saying, “Hey, we’re going to take this one environment and we’re going to try it out on CXD,” and that could be just a few clusters, right? That could be 12, 16 nodes… But we also have SMBs in our datacenters too, and they’re saying they just need a cluster, “We have this one environment, it’s up for a tech refresh, we don’t want to spend the cash and this seems like a better model, and for whatever reason we can’t or don’t want to run it up in public cloud.” So, I think we’re starting small as people get used to us, and I anticipate, just giving what I see in our pipeline that that will grow into that higher end, but I think our sweet spot is going to be in the dozens, if I were to predict kind of long time.
Max: [38.59] We’re talking about this from like public cloud to physical migrations, but doing you know, physical server to virtualized server migrations as well… I think it’s really surprising when you go that process and you watch an enterprise go through that process of just how efficient modern hardware is, when you start virtualizing physical servers onto a platform like VMWare and Nutanix, and just what that compression actually looks like in terms of physical servers. I mean, six servers, you know, six cutting-edge, modern, HCI, Nutanix nodes… I mean, you’re talking hundreds and hundreds of VMs that can come onto these machines –
Holland: [39.30] Hundreds. Absolutely, the consolidation ratio – and obviously it all comes down to the workload type – but likely BDI instances… I mean, you could — on one box you could be talking an 80:1 consolidation ratio, a 100:1 consolidation ratio, so… Yeah, it’s beyond impressive and you can be extremely efficient with how you do those consolidations and with a platform like ours, you don’t have to worry about pinning that infrastructure, either from a compute standpoint or a network standpoint, up in its redzone, because there’s no metering. Another thing I wanted to point out too in terms of efficiency and cost consideration is, if you are generating a big center of gravity, data gravity, on that infrastructure, you don’t have to worry about if one day you want to take it out from Cyxtera and put it somewhere else, there’s no notion of an egress fee with us, as is not true with some of the other platforms. So, as you extrapolate your data usage, down maybe three, five years, you don’t have to have that worry, and I think that is a consideration that you do need to think about with other platforms.
Max: [40.35] So your partnership with Nvidia, and talking about GPU compute, I mean I go to a very specific application list very fast, so… I’m curious what’s driving this for you guys and what you say as like, we need to get into this side of the house?
Holland: [40.51] Yeah, it’s very much where we’re starting in Nvidia and their hardware and software platform has a lot of different applications, as you know, across a lot of different verticals, but the one that we chose to start with was definitely AI and machine learning… We got on this journey with Nvidia last July, I believe it was, by having a handful of our datacenters certified as what they call DGX-ready, it’s one of their programs, and that basically just says that we have the environmentals, the cooling, the power capabilities to support these high-performance compute nodes, right, that might consume many more kilowatts than what’s standard, per rack. And as we went through that certification process and started engaging our customers, we started to see command signals saying, “Gosh, we’re running a lot more GPU enabled compute nodes,” and I go, “Okay, well that’s interesting, what are you doing on them?” Inevitably the conversation went to AI and machine learning, and this has applications everywhere: governmental, fintech, automotive, oil and gas, you name it. And when we saw all the folks that are sitting in our datacenters anyway, so then we started to say, “Okay, well what are some of the challenges to adoption today? Where are people running into headwinds?” And as you know, some of these systems from Nvidia, especially their DGX systems, can cost hundreds of thousands of dollars, so… We said, actually, what if that was available in an on-demand offering, delivered in an automated fashion, on our SDN fabric, and people could subscribe to one and take advantage of those efficiencies of being able to run it unencumbered in terms of resource monitoring and then the associated billing. You know, what would you think about that Mr. Customer, and it was a resounding ‘yeah that sounds really interesting’. So, we’re starting out small-ish, with the Nvidia relationship, we’re targeting some very specific verticals, but really we’re seeing a blossom of AI and related use cases and workloads, and that’s the space that we’re hoping to help our customer with out of the gate, there’s a whole other set of conversations that are likely to occur around gaming and just generally GPU-enabled workloads, which seem to be taking off in a bunch of different directions… But right now, we’re going to try to cut our teeth and get the AI and ML piece right, and then we’ll expand from there.
Max: [43.11] I mean this goes back to you know, financial conversations from two sides, right? Are you investing the capex to buy this or are you currently running this workload someone else? When we talk about public cloud vendors and we look at, you know, actual cost, true cost in public clouds, you know, data egress is a big factor for a lot of people, I mean more than they would expect it to be, but you know, the data science and machine learning pipelines, for companies, I mean these are massive cost centers and cloud environments, I’ve seen lots of companies carve out machine learning, data science, workloads and shop those around and find different providers and CC’ing primary platforms on one cloud and machine learning on another cloud, and I think that’s become a very common kind of expectation of… We’re going to start chunking those things up, and taking that out of cloud entirely and bringing it back to a physical environment that a customer doesn’t have to operate, I mean that’s… That’s a very interesting opportunity for conversation and also a lot of value you’re presenting, because people have this problem now of hundreds and hundreds of thousand dollars a month in spend to maintain these things, that are unpredictable… And cost variable, and… You know, not CFO friendly, right? I mean, that’s — it seems like a really good match for you.
Holland: [44.27] It certainly feels like it, and look, you know, not an acknowledgement to public cloud, but sometimes that’s the right place to start, but then as you start needing more GPU nodes, you need to cluster them together, you start forecasting that data expansion, at some point there’s a strong argument to be made that at least some of this should reside in a dedicated environment, that’s not so metered. But yeah you know, we’re trying to cater to, initially out of the gate, the data science community, but at the end of the day just want to train our models and get their application off the ground, and they’re more than happy to go take the credit card and consume up in public cloud, but at some point the CFO shows up and says, “What’s going on here, this is getting pretty expensive,” and I think we’re going to be a really attractive alternative to that, being able to run at a fraction of the cost.
Max: [45.19] You know, obviously Cyxtera is still very much a datacenter company, and you offer different flavors of that in different markets, and CXD is… I was very excited when I saw this announcement, and it was something I think you bring a lot of value to the market, and we see now more and more conversations — multi-cloud has been in conversation for a long time, but not many people are in multi-cloud and hybrid cloud became a conversation and now we’re starting to see movement into bare metal and it feels like you’re in the same boat when you’re seeing conversation in customers that went from colo to public cloud but now they’re coming back and not wanting to be in colo anymore, and this is like, where you’ve started from and you’re just like, coming full circle back, right? We were colo, we went public cloud, then we’re back colo but now it’s kind of like in the middle, you know? It’s funny to think about in that way.
Holland: [46.06] Yeah, it very much is, and everything comes in cycles, we think we’re well positioned to take advantage of this next cycle, which is very much having a bunch of choices and being able to architect a solution that’s right for your company, right for your workloads, and we feel like that what we’ve created is something that was missing in the market, and at the end of the day we’re just here to give our customers choices.
Max: [46.30] So when you said about your partner ecosystem that is available on CXD, you know, storage was your example, the first one that came up… What else kind of like, categories are available to somebody on the partner platform?
Holland: [46.42] Yeah so, right now we’re, I’d say storage… The compute, obviously, we’ve got a nice ecosystem of a connectivity, whether those are network service providers or SDN providers, we’ve got a great relationship with MegaPort as an example… You know, packet fabric, books of that nature. But in the queue to come and join the CXD party are technology vendors in a number of different categories, so think security, think networking, right? Anything that can be addressable via Layer 2 or Layer 3 that can plug an ethernet cable into has the opportunity to be delivered as a service on our platform. And the end state vision that we’re trying to accomplish is the full ability for you to be able to deploy — say you need to rapidly expand into a new market company, or say you need to set up a new availability zone for your x as a service platform, whatever it is, wouldn’t it be great if you could just go to a portal and start clicking buttons and basically design a datacenter, and then have all of that come to life, all programmed to the API, and be available so you can move your container images or your VMs over there and now all of a sudden you’re up and running. You’ve taken a six month process maybe down to a couple of days. If we do our job right, you’ll be able to do that.
Max: [48.05] So if I have a partner, I have an application or a need for one of your partners, is this something you’re enabling the connectivity to it and I can provision that in a portal, or is this something where I actually have to go out and establish — to bring the lawyers in and go through the MSA and go that way, or is this something like I already have an MSA with Cyxtera for CXD and you’re billing me nad I can go click on a button saying I need this thing, just turn it on, and Cyxtera bills me and I don’t have to deal with that again?
Holland: [48.32] Yeah, yeah… I’ll go back to my wonderful answer starter again – it depends. So we do have some services that we can resell, a small handful that would sit on Cyxtera paper that would save you having to go negotiate another contract, but really what we’re doing behind the scenes is enabling the connectivity – Layer 2 specifically – connectivity out to these ecosystem of partners. So, we would go into our web based command center portal and say like, “I want to connect over to… connectivity provider X, storage provider Y,” and we will basically send a friend request to them, and there’s a little bit of back and forth, pretty simple, but to establish that relationship. We don’t want people to be able to rouguely just attach to each other, so we have some safeguards in that process. But, at the end of that short process, we effectively stretch one of more Layer 2 segments through our CXD fabrics, through these VXLANs that are dedicated just to that customer and that partner, and then they can transact and consume that service in whatever way they want. So, the magic really is the automated connection from A to B, within the facility, and then the commercial relationship that sits on top kind of varies, depending on our relationship with the partner.
Max: [49.45] So infrastructure as a service, IaaS, became a really popular marketing thing and really what that was was some kind of private cloud platform, multi tenant environment that a service provider was managing and providing services for, and you guys are very different because we’re not talking about you providing any services to the customer outside of… This infrastructure is available to you, do whatever you want to with it. So when I think about enterprises and where these, you know, other fits, we talk about existing Cyxtera customers but then we also talk about, you know, if an enterprise has infrastructure in a datacenter today that maybe is already VMware, you know, they’re running VMware or they’re running Cisco’s platform or they’re running different things because they want to have off-site, in a colo, and not go through a refresh cycle in future, I mean Cyxtera would be a really good, interesting option for them because they would still manage it and they would still maintain it, it would just be — that don’t have to support the datacenter, they don’t have to drive bodies there, they get all these ancillary benefits around it on top of your platform. I mean, that’s kind of what I’m hearing you say.
Holland: [50.48] That’s absolutely right, yeah, and it brings up a good point of distinction that I’ll underscore again, which is while we manage the network and we do some management of the hardware, because the compute nodes specifically are sitting in a secure cage that only Cyxtera folks have access to… If a drive goes out, a prop goes bad, memory stick goes back, we’re out doing those break fixings, right? But then we start moving up the stack into the application and the hypervisor layer, whatever, for those customers who don’t want to do that type of thing, you know, by themselves – or with us, because we’re probably there troubleshooting alongside them, because sometimes a hardware issue can manifest as a software issue, so we just don’t know what it is, we’ll be on the phone with them… But a lot of folks do want a managed environment, in fact we’ve got a great relationship with a number of MSPs, all call out serious computer solutions, because they’re one of our first… We partner tightly with them, in fact they build an HCI as a service platform on top of our CXD platform, and so you can go get that full managed experience, all the way up the stack and all the adjacencies, maybe you need a SIM, maybe you need a managed firewall, they do everything and we’re just kind of this partner that’s in the background, silently powering that engagement through the managed partner… But yeah, those folks are critical to us, because we can’t do everything, we’ve chosen not to do everything when it comes to managing these environments. And that neutrality, I think it has served us well, because it doesn’t put us in competition with our various partners.
Max: [52.12] Give me a ballpark – and I’m going to answer a lot of the ‘it depends’ questions here before I ask you the rest of the question, right? So if you think about it from base infrastructure, right? You’re talking about a company that’s doing — either wants to go from physical to virtual or they’re doing a refresh in their existing virtual environment and they know their nodes – let’s just call it three nodes – they need three knows, HCI environment, you know, they’re in a relatively vanilla configuration, dual socket, some number of core, an average amount of RAM… You know, an average amount of SSD or nVME drives… What would they expect, what’s that kind of ballpark range of what they would be looking at to come over into a Cyxtera CXD?
Holland: [52.54] May I ask for how long they would want that environment?
Max: [52.57] Let’s call it three years.
Holland: [52.58] Okay –
Max: [52.59] I mean, it’s enterprise, right? So it’s for three years, right?
Holland: [53.01] Yeah, yeah. I would say… Let’s call it somewhere between – I’m going to give you a midrange here – eight hundred to a thousand bucks a month, something like that?
Max: [53.12] Per node or for the entire caboodle?
Holland: [53.13] Per node, but what’s included with that is all of the networking infrastructure, top of the rack, core everything, that’s all of the… I’ll call it smart band services, which is us doing all that hardware break fix in the background – that’s all the power, cooling, kind of all those other inputs that we might factor in to charging a colocation customer to come and the kind of denizens of our buildings, that’s all included. And I think that power input’s a big thing that I think people mix, because they might say, “Gosh, eight hundred bucks a month, I could go lease this and it would maybe be $650 a month if I did straight lease,” yeah, but what about everything else? What about the upperx switch you’ve got to buy? What about the fact that it’s sucking down five hundred watts or eight hundred watts of energy? I can translate that into real cost and all of a sudden that cost of per node at eight hundred bucks is looking pretty good. So, that’s one of the things that we have to sometimes remind our customers, that there’s a lot built into that node cost, because you’re not having to think about anything, you’re not having to think about building security, you’re not having to think about the fact that we’ve got three layers of power redundancy, fourteen providers of fibre sitting at the edge of the building, and all these things add up to I think a pretty nice value to add and certainly ahrd costs that they’ve incurred that they’ve had to go and solve themselves.
Max: [54.25] So you’re saying eight hundred to a thousand as a rough kind of mid-tier ballpark, right? So we’ll say twenty four hundred to three thousand dollars… When you compare that to a datacenter, after you have cabinet space and power and cross connects and network services and start factor — I mean, just that alone, you know, somewhere in the like eighteen hundred to twenty five hundred dollar a month target, I mean… You’re already there, talking about buying equipment. So that’s… I wouldn’t say that’s — I mean, that’s incredibly competitive, when it comes actually into, what is the total cost of supporting a datacenter environment and not factoring in like, can you have somebody physically local that can drive to the facility and go do, you know, physical maintenance?
Holland: [55.07] Right, and which we can do as well, but yeah, so for that three node — fictional three node cluster at twenty four hundred bucks a month, I think that’s a pretty good deal. And let’s pretend for a second too, and this is a huge differentiator for us compared to almost every other bare metal platform, is that… Let’s say you have a technology stack that needs to sit adjacent to those compute nodes – because we know it takes more than just a compute node to launch a workload, right? You might be a Juniper shopper, a PaloAlto shopper, a Fortninet shopper, and you might want to have their security stack deployed alongside of it. You might have an SD-WAN solution, you might have some exotic networking equipment, maybe you have an old AS400 that needs to interact with this cluster. Whatever it is, we’ve got CXD enabled cages and environments that we can set up that we can pair with those compute nodes too, and it’ll all work together as if it’s sitting right next to each other on the same LAN, effectively, right? But then you still get a lot of the benefits of us being able to do smart hand stuff, we’re happy to come in if you’ve got a red blinking light or you need us to jiggle a cable because something’s not working… That AS400 might be collecting dust, it might need a little… One of those cans of air at the back, or whatever – we can do that sort of stuff too. So you get a little extra possibilities with the platform if you need to pair it with our services, or some gear that you want to send us and have stand up alongside.
Max: [56.25] You know, those AS400s are going to be running until the end of the internet, I don’t think we’re ever going to see the end of those things.
Holland: [56.32] They’re going to outlast you and me, that’s for sure!
Max: [56.35] One hundred percent, so. Well Holland, thank you so much for your time, I’ve really enjoyed this, it’s great.
Holland: [56.42] Likewise Max, this is fun.
Max: [56.44] Yeah, I look forward to doing this again, you know, in the future.
Holland: [56.47] I hope so man, happy to come back any time, I appreciate you having us on.OUTRO: [56.50] Thanks for joining the Tech Deep Dive podcast. At Clarksys we believe tech should make your life better, searching Google is a waste of time, and the right vendor is often one you haven’t heard of before. We can help you buy the right tech for your business, visit us at Clarksys.com to schedule an intro call.