Accelerating Data Center Solution Deployment with Joel Wineland, CTO of Racklive

(Intro Narrator at 00:00:04) Hello, my friends. Today, Joel Beasley is talking with Joel Wineland, CTO of RackLive, and they discuss how deploying fully integrated racks can significantly accelerate deployment. All of this right here, right now on the Modern CTO Podcast.

(Joel Beasley at 00:00:39) This is the Modern CTO Podcast.

(Joel Wineland at 00:00:51) So my story. I don't know how far back we want to go. In college, I was a Luddite and an art major. So my belief was that technology went too wide and too fast and was not an asset to deep thinking, I guess, is the way I would have phrased it. So I wanted to spend my time making oil paintings and charcoal drawings and things like that.

(Joel Wineland at 00:01:11) Ended up getting married while I was in college and needed to get a real job. So I'd seen some ads on the radio, I think. Took some computer certifications. So I started out with a really basic class thinking, alright, cool, I can go get a job for $25 an hour or whatever it would have been at that time. And kind of fell in love with it, realized that the diversity and the uniqueness of what was inside platforms was really very similar to what appealed to me about art.

(Joel Wineland at 00:01:35) So I kind of applied that same mentality to what I was working on from a technology standpoint. Got a bunch more certifications because it was fun to collect them back at that time. Those days were kind of silly, but we were all really proud of our MCSEs and A-pluses and Server-pluses and this and that. So I ended up going to work for Dell. At Dell, I had worked in tech support for a period of time and then got promoted to work in the storage area networking design team.

(Joel Wineland at 00:02:00) That led into Dell Professional Services, and I got embedded at Walmart for, I guess, close to four years. At Walmart, I did a turn through Dell Professional Services and then through Advanced System Group. Ended up doing a lot of different design work with them on platforms and then a lot of server deployment, so I got my hands dirty, which is part of what we like to do here as well. From Walmart, moved on to Rackspace. They kind of came through.

(Joel Wineland at 00:02:21) I had a buddy who worked there and ended up recruiting me. Moved in there as a product developer. So my focus from product development was hardware. So I kind of went in there, and originally, we were buying kind of all package systems from OEMs. So Dell and HP were kind of the two primary that we were running for the most part.

(Joel Wineland at 00:02:37) But the initiative we were driving there is we were kind of formulating the cloud, so our precursors to what was Swift and what became OpenStack. So ultimately ended up kind of running platform design. Initially, just me and another guy, kind of grew into a group of about eight people where we were kind of evaluating and kind of factoring in what was needed to deploy server platforms that were more specific in their design rather than kind of accepting what was framed up by the OEMs. That led into kind of our Open Compute pursuits, which was a lot of fun. So that was my project for about two years there.

(Joel Wineland at 00:03:08) I was sitting on the OCP Incubation Committee and kind of running platform design around what we needed and how our racks were a little bit different than what Facebook and others would require. Really, really loved doing that. But ultimately, I realized that while Open Compute was really exciting and amazing for someone like a Rackspace or Facebook, it was very hard for smaller consumers to get their hands on to kind of get in and consume. The other thing that we'd done while I was there was done our first full rack deployments. When I arrived, everything was deployed out of cardboard.

(Joel Wineland at 00:03:39) So we were literally unpacking everything, pre-staging racks in the data center and hand-installing all that. The data centers hated us because we had piles and piles and piles of cardboard for deploying tens of thousands of servers, right? So it was harder than I thought it would be, to tell you the truth. Originally, I thought, well, what's the difference? Just put all this stuff in the rack and lace it up. But the lessons learned there were consistency is king, right? You really got to be able to button things down, document what you've got, and do things repetitively very, very accurately. Otherwise, rolling things in the rack gets to be a lot bigger mess than it does just individually installing them.

(Joel Wineland at 00:04:14) And then secondarily, sometimes you can go beyond what's necessary in a design and make your life a lot more complex. For instance, I remember one story where we were deploying something like 30 cabinets, and we needed to pull five cabinets and deploy them to another room for a different solution. But because of the labeling, this is the only difference between the systems and anything else in those cabinets, they were labeled differently. We ended up having to fly to Chicago with sheets of labels and relabel hundreds and hundreds of cables on both ends just to go ahead and be able to roll that rack into a different spot on the floor. So anyway, after I left, I joined Rourke Data, which was a division of Avnet at the time, with the goal of trying to make Open Compute easily consumable, get it on the ground, get it in the warehouse, get it on the dock. Had some great fun there, but realized that ODM design is a lot of fun but requires a certain lift that's not necessarily for smaller consumers.

(Joel Wineland at 00:05:09) If you're deploying one to three racks, at least at that time, it was very hard to get your hands on it because you didn't have any clarity on what was actually coming out the other end of the supply chain. So it was really incumbent upon the large consumers—so Facebook, Riot Games, Rackspace, and others—to kind of speculate and define what was needed, and then we changed pretty fast. Because as our needs evolved, we'd go ahead and add, change the resources. So each large-scale production run started producing things that were somewhat different in terms of their feature set and their capabilities, where we were trying on the other side of things to get things packaged and available and in stock so somebody could order some and have them on the dock within a month. So realized there that there was something different that was needed.

(Joel Wineland at 00:05:48) So ended up having a couple of wins with the team here at RackLive, and so they said, "Hey, can we come in and kind of reframe this? Think a little bit differently about what the hardware looks like and how we can help people amplify what they're actually deploying." So taking a look at that, the realization was that experience building racks was something that a lot of the consumers that they had worked with historically were not really that familiar with.

(Joel Wineland at 00:06:10) So their kind of knowledge of how to craft a standardized template, how to, what to put in, what not to include so that you wouldn't be so specific that you limit your flexibility, but so that you can still support it and still get it laced, clean, and deployed. So we built this practice, and actually, the team here had started about three years prior to me joining with a bunch of guys who left Rackable Systems. So they had a strong background and a lot of knowledge in deploying racks. But the trade-off for us, and I think the thing that makes us unique in the market, is we're very agile and dynamic. I don't know who I can name in terms of references, but we've had some social media video companies who originally had a single template that we diverged into three that then, ultimately, we were changing the dynamic for every pop that they were deploying around the world. So I did 23 countries with them with nearly 23 different designs.

(Joel Wineland at 00:07:01) I think it was more like 18. But, so what we realized there was we've got to get the documentation, the standard templates, and the test elements kind of framed up so we can move really fast, so we can test things in a dynamic way, but not be so locked in that when they say, "Hey, I've got to change my tors for this deployment. I've got to move to," that time, it would have been one gig to 10 gig, that we're so locked in that it takes months to change the MPI and the build documentation and the structure to go ahead and get those things laced and integrated to their spec.

(Joel Wineland at 00:07:29) So anyway, that's kind of what led us here and my backstory in a way.

(Joel Beasley at 00:07:33) Nice. So when I go home tonight and talk to my wife, I'm like, "Hey, I talked to this guy, another Joel. Very cool. He's at RackLive." RackLive does this. What is the this?

(Joel Wineland at 00:07:43) Sure. So our claim to fame is really a fully integrated laced rack to deploy into a data center. So basically, all you've got to do is go ahead and drop your power whips below the floor, throw your uplinks up, and the rack's off and running. Depending on the customer, we'll have a variety of different insertion points into the application stack. Often, we're deploying the operating system, preconfiguring things like their IP addresses, host names, SSH keys.

(Joel Wineland at 00:08:05) To a certain degree, applications, although most customers today prefer to deploy applications using some orchestration platform and don't require us to do a lot of that just to get the basic framework up and running so when the rack hits the floor, it can ping.

(Joel Beasley at 00:08:18) So it's like a fully enclosed system? Like, you completely assemble the rack and everything. You ship it in like a giant wood crate or something, and you just take it out?

(Joel Wineland at 00:08:25) Yeah. Yeah.

(Joel Beasley at 00:08:26) Oh, that's pretty cool.

(Joel Wineland at 00:08:28) Yeah. You can kind of see this is a vinyl rack behind me that represents pictures of some of our racks, and that's exactly what it is. It's fully enclosed, fully wrapped up. There's a lot of kind of diversity and complexity in terms of how things are laced and deployed. For instance, you'll find a lot of networking gear and other equipment that has different directions of airflow.

(Joel Wineland at 00:08:46) So we've got some racks where I have front to back, back to front, side to side, and side to front, for instance. So ultimately, we have to do a lot of things to go ahead and baffle and design that rack to handle airflow so that you don't end up mixing hot aisle into your cold aisle or confusing the scenario and compromising your efficiency. So yeah, that's what we build. You can think of it as basically what ends up in the data center gets built right here, gets pushed into a giant wooden crate, gets forklifted into trucks, and shipped all over the world.

(Joel Beasley at 00:09:12) That's awesome. I got to do some touring of my county's emergency operations command centers and things of that nature, and I got to go visit their data centers and meet different technology leaders. And they have whole contractors doing all of this work to assemble these racks. Because I remember being there, and they were doing some upgrade, and there were all of these different types of people involved.

(Joel Beasley at 00:09:35) And the idea that you can just have a done-for-you service, you just call up Joel and be like, "Hey, we need these racks," and you guys just do it. That sounds awesome. So like, with the world going to cloud, right, who are most of your companies? Like health care companies? Who, where do you fit in the market?

(Joel Wineland at 00:09:52) It's an interesting question. So we don't do a lot of marketing. We're more kind of tightly entwined with those that we know, and the guys that we work with kind of move shop to shop to shop. So it's a lot of word of mouth, and we've been around—ASA is the parent company—it's been around for 35 years.

(Joel Wineland at 00:10:05) RackLive now has been around for going on 12. So functionally, what we end up looking at is really a large swath of advertising technology. So we work with a lot of the people in the ad tech space, both because they have a lot of geodiversity. They're trying to deploy kind of pops and elements all around the world at typically colocation facilities.

(Joel Beasley at 00:10:25) Can I interrupt you? What is a pop? I don't know what a pop is.

(Joel Wineland at 00:10:28) Point of presence. So a small or smallish footprint might be one to five racks that's deployed in a location that's often kind of near your point of consumption or your user base.

(Joel Beasley at 00:10:38) Oh, so if I was like CloudFlare, maybe that I would need different places based off of how my business is growing in different locations.

(Joel Wineland at 00:10:48) Absolutely. Yeah.

(Joel Beasley at 00:10:49) Sorry. I like to interrupt because when I hear these things, I don't think he's talking about the email server. So I did have a question. When you were doing the background stuff, I wrote it down.

(Joel Beasley at 00:11:04) You were at Dell, but you were like an employee that was inside of Walmart, like the sort of employee leasing type deal for lack of a better term. And so my question there is, I have noticed a trend of people coming out of Walmart that are awesome technology people. Everyone from like Jeremy, I think he just moved over to be the CTO of Pinterest, to Asir, I think, and he was like CTO of Williams-Sonoma, just really great people. And so my question to you is, there's obviously this huge company, many different departments. Did you pick up on that too? Like, did they have a great culture when you were there?

(Joel Wineland at 00:11:45) So yeah, they had a very tight culture. And at the time, the diversity of deployments and scale was something that was kind of pre-cloud, to be honest with you. I think you saw Amazon and others kind of growing up and around them, but it wasn't common to see a lot of people consuming IT from a utility standpoint with very large environments. So you could kind of view basically what we worked in there as a precursor for cloud, a very, very large set of enterprise data centers within a vast distribution with thousands of stores with their own kind of data center footprint and other equipment that's embedded in them. So ultimately, I think just the problem set that they had to solve kind of made some of the best technologists gravitate toward that space.

(Joel Wineland at 00:12:22) And yeah, some of the guys I got to work with there, or some of the team I should say I got to work with there, were literally amazing. So a few of them kind of left. One came with me to Rackspace, ended up being a CTO for another shop in Canada, and now is working for a much different place trying to change the way people consume. How do I say this? Anyway, working in the medicinal products industry.

(Joel Beasley at 00:12:48) I have no idea what you're talking about. Yeah.

(Joel Wineland at 00:12:55) Oh, that's great. Yeah, early on, I had this really cool guy. When we first started the podcast, name was Roger, and I always butcher his last name, like Obonado or something. But when that space was expanding, like, when they were getting the first approvals, all the credit card processors backed out, right? And so he came in with a point-of-sale system and a processing system, and his business just exploded. And so I got to see him, meet with him right when he was in the middle of it, and then watch him. And now I have mutual friends with him, watch him as he sold the company and is now off advising other companies. But I was like, man, what an interesting spot. What a great position to be in because it's not that controversial doing the point-of-sale systems, right? So I was like, dude, good job, Roger.

(Joel Beasley at 00:13:39) QTS Data Centers. I know David over there who's a super awesome guy, and they mentioned that they're one of the largest by square foot data centers in the world. Are those companies, do they have their own internal teams that are doing this? Are they working with contractors like you to help when they boot up new facilities? Is there any connection there?

(Joel Wineland at 00:13:58) So typically, yeah, they'd work often—and I don't know QTS directly—but they'd work often with people like us to deploy any infrastructure that they need. So it depends on the scale of the infra. A lot of guys would go ahead and rack and deploy their own gear to a point. But if you're going to need 30 racks or more, ultimately, that becomes a huge headache and is a huge mess.

(Joel Wineland at 00:14:16) Plus, it also depends on the dynamism, right? How many racks are you trying to deploy? If you need kind of a pre-staged thing that gets you some storage capacity or whatever up and rocking and rolling, but as you're deploying that kind of cyclically, you know, month over month, three cabinets or six cabinets to keep things growing out ahead, ultimately, that becomes a huge headache.

(Joel Wineland at 00:14:33) I think in general, though, QTS, I'd view them as more of a colocation facility with the capability to offer advanced services, and 90% of what we deploy ends up in colocation facilities that's sourced for our customer base. So their customers are our customers, and basically, we're partners in that scenario.

(Joel Beasley at 00:14:49) One thing that caught my eye when I was doing the prep for this conversation is a couple times this thing was mentioned called the circular economy. I was like, I've never heard that before. I was hoping you could explain it to me.

(Joel Wineland at 00:15:02) Sure. Sure. So that lives more in the realm of IT Renew and other partners who are cycling things that have reached their end of life for a given purpose or maybe exceeded the value for a particular application set but haven't exhausted their financial potential value. So in the circular economy, what you're looking at, and this is now a kind of top-level open compute project and has a lot of champions, is you're looking at people who purchase—I don't want to say used gear, but maybe long-term burned-in gear if you want to think of it in those terms—and then refactor those so that they're usable for different purposes and then cycle them out. So you can think of it basically as we see this cadence of technology kind of driving us forward. In my opinion, we're in the most exciting time from a hardware standpoint because what we're seeing now is computing in general has been kind of viewed as commodity, right? You have cloud that offers you a very diverse set of resources. So ultimately, when you want to go ahead and deploy your own gear, you either have a need to drive some specialized set of capabilities, reach maybe a geo that's something that's not served very well by the cloud providers, which is getting narrower and narrower all the time, or, of course, cost, which can play a big factor, and that's why we see a lot of our customers repatriating from the cloud. But secondarily, as you look at that bulk computing, you start to think in terms of, all right, cool, the computers that I deployed three years ago, right, they're depreciated now. That asset's basically time to be cycled out and taken off the books. But it's not a non-workable entity, especially as I'm looking at, you know, microservices or virtual workloads where the hardware specificity may not be that dramatic. So what you're seeing in the circular economy is both the opportunity to kind of change the cost constructs that are driving hardware acquisitions by using what, you know, I don't want to say used, but using used gear, previously deployed gear maybe is a better term. And then also, you're looking at people who have a new type of environmental sensibility. This is something I didn't know until I started working with my friends at IT Renew for a while was basically, when you look at what's happening in the industry, most of the embodied carbon that gets into a platform really comes prior to that platform mining. It comes from shipping. It comes from manufacturing. So ultimately, by preventing the need to have new systems manufactured, you can really allay a huge benefit to the earth overall by preventing some of those upfront costs and demands for some aspects of workload. So it's typically not a space that Racklive spends a lot of time. We're pretty aggressively charging upfront. But I think, in general, it's a space that I'm very supportive of. It's one that I think makes a lot of sense if it's a workload that fits that use case.

(Joel Beasley at 00:17:32) Yeah. Thanks for explaining that to me. Quantum computing, as you were talking, I was thinking about this. When you see emerging technologies like that, right, what are your thoughts on it? Do you start offering it? Do you wait for your customers? Like, how did quantum computing come into your life, and does it apply to Racklive?

(Joel Wineland at 00:17:51) Good question. I mean, I'll answer it easily. Not today. No. Our customers, this is something that when I kind of started my journey into making open compute more generally consumable when I left Rackspace and tried to kind of bring this more customized and tailored kind of solution mindset to a much broader user base, I realized that there's a number of dynamics that play into deploying things that make sense at a certain scale, but maybe don't at a much smaller or a more transactional kind of scale. So I'd say quantum computing is interesting to us, but isn't part of anything that we're leveraging today. Where we're seeing the kind of bleeding edge from our perspective technology come into play is something that's been around for years, right? I mean, using GPUs for general-purpose computing is not a new capability set or not a new resource. Now they're seeing more frameworks made to consume those. Look at Kinetica or Spark Rapids and other kind of package frameworks that can layer on top and actually make database consumption on a GPU-accelerated back end something that is very easy, or I shouldn't say easy, but it's very direct to adopt for application developers. We're starting to see that metamorphose, and that's kind of the area where I think the most excitement in the industry comes from is application of what you'd call AI/ML workloads to problem sets that traditionally were kind of out of scope, whether that's optimizing your environment and deploying workloads where they're best suited or whether that's actually the core aspects of what you're doing in terms of image extraction, feature extraction, natural language processing, et cetera. So that's where I see the most excitement where we're at today. Moving into alternative computing architectures, we are seeing finally, and this has been something I've worked on since I was at Rackspace, the attempt to move Arm into the server arena to make it a realistic server platform. So worked with a number of different companies who had kind of really interesting, I'd say, almost false starts in the space. Did pretty well. But finally, we're now seeing some traction in that arena with Ampere and seeing the Arm platform start to become a realistic contender for computing workloads that a lot of our customers consume as well. But I think a lot of our guys are kind of rolling with the punches in terms of their apps. They're growing fast. Typically, we're with people who are either taking their first deployments off of cloud that were with us or kind of expanding that footprint. And I don't want to say that our customers aren't using cloud. They certainly are. Everyone does. But our customers have identified that area where it's either business differentiating to go ahead and deploy systems into a more diverse set of locales or where they can differentiate, whether it's cost or capability, by being more specific on the hardware they're deploying and deploying to the now. So that's kind of where we stand.

(Joel Beasley at 00:20:23) You're saying you're excited because they're going to bring Arm to the cloud?

(Joel Wineland at 00:20:26) Arm's already heavily in the cloud. I'm excited starting to see my customer base willing to adopt Arm and starting to deploy it for some of their core workloads. I think having diversity in the computing arena is excellent, and Arm gives us a lot of really unique capabilities and having a tailorable architecture that really is kind of metamorphosed with the use case. So it's not that I'm not an Intel fan. I definitely am, and the largest majority of what we deploy is Intel today. But I'm starting to see some traction and more diversity where we'll see both the competitive pressures add some value, and we'll also see that people can start to do some new things and think differently about what defines a hardware platform than they did, you know, two years ago.

(Joel Beasley at 00:21:06) For the supply chain stuff, I'm kind of a nerd, right? And when I saw those ships out there and everything, my first thought was—and I'm bringing this up because you like automation—I was like, why is there not automation deployed here?

(Joel Wineland at 00:21:19) I agree with you. I think it's a human issue and something that is relatively easy to automate in a way.

(Joel Beasley at 00:21:25) So you're one of the founders of this company?

(Joel Wineland at 00:21:27) So no. Actually, Racklive was founded before I joined. So I joined about three years after they had kind of started to do their initial deployments.

(Joel Beasley at 00:21:35) Yeah. When I saw that you spent some time at Rackspace or when I saw you were in San Antonio and you also were doing, you know, I learned what Racklive was. I was like, oh, I wonder if he was at Rackspace when he's in San Antonio, because I became a customer of theirs at one point, like, in 2007 maybe or between 2007, 2009, right? It's hard to know. But they had just, like, purchased an empty mall that had become defunct—

(Joel Wineland at 00:22:00) I worked there. I worked there.

(Joel Beasley at 00:22:02) Yeah. Yeah. And so I was like learning all about them, and I remember like, you know, the whole process of booting up servers. And then I remember the first time somebody showed me like Amazon Web Services, and my mind was just blown. I was like, this is incredible. I don't have to submit tickets, make purchase orders, get into like long contracts. Like, I don't have to do any of this stuff. And so I was just super excited. And then I was more excited too to learn that like their software advanced enough to where when people buy racks like yours, they can have that sort of cloud experience just on-prem. And when that happened, I think a lot of people were seduced by the ease of deploying applications and that whole, you know, system and process of the cloud. And so they went over there, but then some of them, you know, didn't like it. But when that cloud-like deployment technology transferred over to the racks, I think a lot of people were super excited about that.

(Joel Wineland at 00:22:56) Yeah. Yeah. Honestly. I mean, and you see it in a lot of spaces, right? It's interesting. For us, it's largely very heavy, very disk storage-dense racks within a proliferation of kind of more application-specific, often, you know, GPU nodes and et cetera that are not always, say, subordinate, but are far fewer than what we see in terms of the storage capacity. But if you look more broadly at kind of that automated rack deployment space where you see people like Nutanix and others kind of offering or, you know, Dell's VxRail, for instance, you see something that offers that cloud experience, but you can deploy it wherever you like or actually kind of consume it in a hybrid fashion. It's exciting technology, honestly. We hang on for dear life deploying deep storage racks and their associated infrastructure. But I think in general, I'd see a lot of spaces in the industry asked about health care and other arenas where the general-purpose nature of a densely virtualized environment that's deployed with something like a VxRail can ultimately change that dynamic and give them the same fluidity and flexibility the cloud offers that they can have in-house. So it's exciting stuff.

(Joel Beasley at 00:24:02) Do companies have like split stuff? Because I'm thinking like if, as you know, let's say my company gets super big, right? Like, doing a hundred million dollars, you know, whatever. Got a lot of people. I'd imagine like most of my stuff would be cloud, but there would probably be like 20%, like, my R&D type stuff, the stuff that I don't even want there to be a chance it's on anyone else's platform. And I'd probably have like some racks of my own. Do companies do that, or am I a crazy person?

(Joel Wineland at 00:24:26) Oh, not at all. No. I think it's actually potentially the other way around. What we see generally is everybody uses cloud. We used to say this at Rackspace all the time is, you know, cloud is for everyone, not for every workload, right? So functionally, every customer that I work with uses cloud. Where I see it is either more in developing applications. So it's often kind of as they're getting off the ground, getting things going, because you may not know the growth question. You may not have a predictable kind of curve in terms of how you need to adopt instances or whatever the resource set that you're deploying is. As that grows, the cost of cloud also amplifies, right? So you have transit fees. You have bandwidth costs. You have, obviously, the cost of the instances that you have, which there are various ways to offset. But ultimately, in the end, at a certain set point, my belief is 100% that it is cheaper to run your own infrastructure than it is cloud. Now that's a pretty high set point, right? But you can save money by running your own infrastructure. So what I see a lot of customers doing is getting their application stack off the ground, getting their business running, getting things going using the cloud. And then as they start to grow, they start to go ahead and compartmentalize aspects of that and say, I want to move my bulk storage away from the cloud. I want to move my big data analytics environment away from the cloud. Or potentially use them in concert, so I keep the cloud running for maybe burst workloads or transactional things that come up during the holiday seasons. But I have the anchor of the business that runs in the geos that I've designated that is running on my own dedicated gear. So for us, it's pretty common, and I'd say every customer that we have uses both. Just we're definitely able to keep busy running the infrastructure that they're deploying into their data center facility.

(Joel Beasley at 00:25:56) Do you have like a just-in-time kind of concept when you're ordering equipment or building racks, or do you have a bunch of inventory like in your storage?

(Joel Wineland at 00:26:05) We're pretty much just-in-time. Some of the changes for us in terms of COVID, we've had to take kind of a stand around things like memory and SSDs just because the lead times are prohibitive around some of those aspects. But in general, we're built to order, and our supply chain is pretty much entirely just-in-time minus a few components that are different now than they were two years ago, two and a half years ago. Because we're just-in-time, getting it built, getting it shipped just as fast as we can. So it's a lot of effort to get a three to five rack pod deployed in 60 to 80 days, especially now with supply chain the way it is. So it's a hard run to the finish and get it out the door.

(Joel Beasley at 00:26:45) What about, like, did you experience—because I know you're obviously a nerd too, right? So you're researching stuff. You're understanding the supply chain. One thing I hadn't completely wrapped my mind around is the chip shortage and the cars. Like, it's really hard to get these cars. Can you—did you—have you researched that at all?

(Joel Wineland at 00:27:04) Not in depth. I mean, my read on it personally is that what we're seeing is the kind of transitory impact of two things. One, the demand increasing rather than going down, right? And then secondarily, the impact on the workforce being able to produce. And that production impact is something that tails the demand pretty considerably, right? Because it takes a long time to manufacture a chip and get it into the supply chain so it can be put into a final product. So I think, ultimately, my read on it is we're still seeing the blowback. Well, it's a variety of things because supply chain, like you said, it's not just about chips stuck in the harbor.

(Joel Wineland at 00:27:36) In fact, not even the most significant impacts on our supply chain are about being able to unload chips. It's more about being able to get the raw materials necessary to produce the resource that we require. So my read on that, and again, I wouldn't want to say I'm an expert on it, is what we're seeing there is an impact on the workforce that trickles its way all the way out to us. And the expectation probably, because I think we all had this perspective when things started out with COVID, that as things locked down, demand would tail off. So, ultimately, production could reduce. And now the production did not — actually increased pretty considerably across the board for anything with a chip installed.

(Joel Wineland at 00:28:12) Now we're left with a backlog that's very, very hard to fill. I think, plus, we're seeing some transitions where demand is in pockets where it may not have been expected, or if it was expected, it wasn't accounted for. We're seeing this with GPUs. We're seeing this, you mentioned, with truck chips, for instance. Right? I don't think you can buy a Chevy Silverado right now if you wanted to. Right? So—

(Joel Beasley at 00:28:34) I have one, but yeah, I had to order it. My friend owns this Chevy dealership, and I was trying to talk with him about it, but most of the information he gets is from corporate. Right? And so when I bought it, there's none on the lots. They have no inventory. And when you buy them, you just have to buy the ones that are being shipped in without seeing it. And then, I was so glad I bought it when I did, because I bought it earlier this year. And so it's a 2021. And in '22, they stopped shipping a lot of them with the heat and the seats because of the chips. So they have a little credit to retrofit it. So they just said, "Alright, forget it. We're just not gonna put chips in the seats. We're gonna just ship it without some features and send it off." They, like, the auto start, a couple different things. They just cut. They just cut from the year, and they said, "We'll give you a credit and a voucher, and you can come back when we have the chips." But one of the things I was thinking about was, alright, I've got to interview this cool guy named Jonathan. He's a company called Grok. They make chips, incredibly fast chips that are specific for — I don't want to say what it is. I forget. But they're specific for some task. Right? And so, you know, he did a startup. He made chips. He's out there in the market selling these chips. You know? And if I remember correctly, they weren't super dependent on outside sources. Like, they can continue to make their chips. And I'm thinking to myself, okay, we've been hearing about this chip shortage for at least two years. Where's the company that's just like, "Oh, we need these chips for these cars. Let's just go make them."

(Joel Wineland at 00:30:05) Yep. Well, you know, I'm no expert on the back end there, but to me, the foundry is the problem. Right? And there's a pretty significant lift and a lot of time necessary to go ahead and get a fabricator off the ground who can actually bring the silicon into a usable state. Right? So I think, structurally, what we're looking at there is there's a lot of factories that I've heard are under development, but you're looking at hundreds of millions of dollars of investment to go ahead and bring something up. And then transactional foundries will go ahead and run chips for you or for me if I have an executable design. Even notwithstanding the technology pursuit, notwithstanding validation, getting my chip to work, et cetera, assuming that it is reasonable that someone could produce one, ultimately ends up saying I'm now in a line with a whole bunch of other people who are trying to go ahead and consume time on that foundry to get my silicon actually cut. So I think that — we don't have enough foundries is, in my opinion, the problem.

(Joel Beasley at 00:31:01) But you're right, because I got to talk with Lenovo, the CTO of Lenovo, right when this whole thing happened because, you know, you just brought up something that I had forgotten. It's not just the chips — because it's silicon, right? It's not just the chips for the cars. We went remote. So there was millions to billions of these devices that needed to be purchased, and I got to talk to him about the mistakes and how hard that — yeah, exactly. Got to talk to him about the mistakes and how hard all that was. You know, but yeah, that completely makes sense. If the foundry is agnostic really to the overall thing, then that can be a bottleneck point.

(Joel Wineland at 00:31:41) Yep. I think that's really where our chip shortage comes from is the fact that we don't have enough people who are capable of pressing it, and then there's not enough agnostic ones who can as well. Right? So I mean, even Intel takes up a large amount of, say, TSMC or other foundry's time making things. I've read, for instance, that we're more than a million units backlog on one gig ICs. So that we may have to start to see motherboard designs changing to move away from using that one gig chip for the BMC, the baseboard management controller. So there's availability of other chips, but that chip being so far backlogged now will start to trickle in and impact our motherboard supply or change design. So it's gonna be an interesting next couple of years because I think a lot of this just wasn't foreseen in terms of where the demand comes from. Everyone thinks one gig? Why do I care about a one gig NIC? For Adam, run a 100 gig and 25 gig. But in the end, every system that they deploy typically has that on it just for sideband or for remote management. So if you can't get that, you're either dropping that feature set from your board, which is a design change. You're waiting to go ahead and get it in there, or the design has to kick up and implement something that's really not needed for the functionality, but just needed because it meets the needs of someone trying to consume it, and it has to be lit up and has to run. Right?

(Joel Beasley at 00:32:54) I've got a couple more questions. Does anyone ever request crazy racks, like, that you design them or color them or put some print on them, make them look interesting? Has anyone ever done that? Oh, yeah. Yeah. Yeah.

(Joel Wineland at 00:33:04) Yeah. So, you know, we actually really enjoy doing that kind of stuff. So we have one very large customer who's, I guess, in the Web 2.0 — I don't know how to describe them. I can't say any names that I'm not allowed to say. So anyway, but basically, they have their own kind of iconography. So their own characters, et cetera, that they use very prolifically, very significantly. So they had us do printed sides for the rack, silk screen with these characters doing different things. And then basically, a lot of times, we're deploying rows of gear. Right? So you'll start in with your first three racks and then move on. So for theirs, basically, we're skinning the ends with modular panels. As they'd add their next ones that come in, snap those panels off, move them to the next end. There's a pretty slick, pretty slick look. I have some who want lights in the rack to go ahead and illuminate certain bits. You'll see this with, just as an example, people like Cerebras or Samba Nova, where they're trying to illustrate that they have kind of new, kind of key differentiated technology. I don't want to say necessarily that either of them would be doing that, but I think you'll see more differentiated bits like that. And then also see some — you know, we do a lot of custom rack design because we have our own in-house mechanical engineering team. And, usually, that's bracketry to go ahead and handle support for additional equipment, to let me deploy a 4,000-pound rack that you can move across the floor. You have to reinforce the floor, typically, but to let you deploy, say, 600 or 800 hard drives in a rack. The other things I've seen is in some of the crypto space, we have guys wanting racks that are three high. So you forklift one rack on top of the next, on top of the next, and bolt them together. Now that's a unique problem, and I don't want to say that we're not — we've solved that one yet. Ultimately, yeah, we see a lot of unique customization, whether it's cosmetic with cartoons and things like that, or whether it's kind of functional and structural like the crazy tower thing.

(Joel Beasley at 00:34:49) That's pretty cool. Yeah. It's like to think of how difficult that would be. Now you're gonna have to build a ladder on the side of it if you need to go up there and make a change. That sounds pretty, pretty amazing. Now I don't have much experience in the industrial side of things, but we did make — like, my wife and I made some friends with some other kids at our church. And I went to see his business and he has — like, they make off-road parts. Right? So he's got this giant warehouse and these machines that are 20, 30 feet long, 5, 8, 10 feet high, and they move so fast, and you can put a metal in there and cut brackets. And, you know, they do off-road type stuff. Yep. And I was blown away by how fast and large these machines were to cut their own parts. Do you guys have any of that, or do you use a partner when you do your AutoCAD designs and need brackets?

(Joel Wineland at 00:35:38) Oh, gotcha. Yeah. We use contract manufacturers for that. We don't have any metal fabrication in-house. We do our own design end to end. So our racks are our rack, and they're about 90% of the racks that we deploy. But, yeah, our CMs are either in Southern California, China, and then one on the East Coast that we typically work with. They have exactly what you've described. Right? Water jets and CNC and plasma cutters and, yeah, impressive stuff.

(Joel Beasley at 00:36:01) Yeah. I saw one of the plasma cutter things, and I was like, "That's a laser?" And he was explaining to me how your building has to be in a specific place for this three-cycle electricity. Like, there's some type of electricity they need in order to do it — three-phase or something. But it was so interesting because I didn't realize that these things could be so large and move so fast and cut with such incredible precision. Everything just seems to be moving so fast in every little corner of our world. And so that brings me to the question about the future. For your area, you're obviously very narrowly — you're an expert in your area. What are some of the things that we're gonna see in the future? And you can choose the distance of future. You can go a hundred years. You can go one year. I don't know, whatever is relevant to you.

(Joel Wineland at 00:36:44) Well, I think in the short-term future, we're gonna see more complex workloads adopted by a broader set of customers. And I think this is something that, you know, for instance, at Rackspace, one of my early projects was evaluation of what we could do with FPGAs. And we were trying to apply that to a couple of different aspects involved with the cloud and network processing. But it ended up that for us, the lift to go ahead and build the team and et cetera to get to where that made sense was beyond kind of the perspective that — or beyond the investment, I guess, that we were willing to make at the time. That's for now, and I have SSDs with FPGAs bolted on that the manufacturer of the SSD provides the capability to go ahead and iterate the code base for that dynamically. So they can literally evolve what the SSD controller can do in-line and change its functionality, give it new capabilities and new resource sets. You've seen this with, say, programmable networking for a long time. I don't think we're even close to the cusp of what that entails. But to make those things generally consumable, and this was kind of my — you know, where I ran my head into, is it FPGA or I consume someone else's ASIC, right, was the lift. Right? If I'm building out a facility that I could save, I could say, "Alright, I save 10 watts per every server, and I can go ahead and reach the same workload." At some point, multiplying that out, I'm building a new facility to account for that power demand. Right? If I'm building a new facility, I'm expending a couple $100 million, depending on how I'm deploying a facility. It's worth it for me to go ahead and make the investment and draw that through. The skill set, of course, has to exist. And I think this is what I'm starting to see now is the commoditization of that skill set, the breadth of it. So moving away from something that's so special-purpose and so tailored that it takes many, many months to reach a value prop from a given resource set or a technology base to now being able to just adopt things off the shelf with some standard APIs, with some capability, evolve those to meet your needs. Now I'm really starting to see what I personally believe is sort of a renaissance in hardware design. So opening up the space that says there's not — it's not like these problems are new problems, really. It's now there's a new accessibility to attack that problem in a different way. Rather than go run out an ASIC that you have to design, tape out, print, and cut, and then wait for a few years to fully monetize, you now really start to see that change come and see new designs iterate rapidly even as we're deploying them. Right? So I think that's where I'd mention things like, you know, Connectic, Spark Rapids, and some of the other suppliers we're working with on different component sets. But I think as you start to look at that, opening up new programmable vehicles and making them standardized, whether through a kind of canonical API or whether through just a kind of, you know, known application packaged use case, is something that we'll really see change things in the course of the next few years. Going out further, you asked about quantum or wet or bio. To me, all of those things kind of live on the horizon that I'm sure when we look at cloud providers who are spending billions, literally, of dollars on infrastructure alone, they're able to go ahead and put the investment in place to look at fundamentally different modes of computation. I think for our use cases and our customer base, we're still a decade away. Right? We'll trail seeing most cars on the road being driven with standard operating systems and et cetera. But I am starting to see a willingness to explore things that are outside of the common mode of deployment. Microservices kind of opened this up. So while most people I work with today still are deploying a standard kind of basic environment — we do almost all Linux — but deploying a standard basic environment and then running microservice workloads or other applications on top of that. I think what we think of as an operating system ultimately will start to be kind of redefined and refactored just because, you know, you look at things like Talos as an interesting kind of adoption in that space, which is a microservices framework and operating system framed up specifically for that purpose. At some point, I think we'll start to see things that are taking that alignment and orientation to build something that's general-purpose and generally capable and really kind of focusing that in, dotting the I's around something that's tailored very well for what it's built for. And it'll change the way that we operate. I mean, I'm kind of a sysadmin and hardware builder at heart, and we've been fortunate in that space by — you know, you know how to punch keys, you know a little Python. You can do some amazing things. But I think that may be changing, and it's time for us to start to think differently in terms of how do we program, what do we program to, and what resources do we use. Because I don't know that we'll expect to see systems that operate like the systems of yesterday in the next five years.

(Joel Beasley at 00:41:09) Well, thank you for that. I use this as investment advice. No. But I like getting to learn about the future and where things are headed from people that are really in it. I've got two more questions. I know we're coming up on time. Do you have — can we do two more questions? Sure. Is that okay? Alright.

(Joel Beasley at 00:41:26) Cool. The first one is, how can I refer you? So when people approach me, what would be the problem that they're having? Would it be that someone approaches me saying, "Oh, I need racks or something like that," I'll be like, "Oh yeah, you should go talk to Joel."

(Joel Beasley at 00:41:40) Or if they have some specific problem and need some specific type of rack, what should indicate that I send people to you?

(Joel Wineland at 00:41:47) Oh, sure. Sure. So I think taking it one level lower, "The cloud's too expensive. I'm struggling with the cost." That's kind of the space where we see a lot of transition and a lot of change there.

(Joel Wineland at 00:41:58) Moving that into our arena, it would be, "I'm thinking about my first hardware deployments, but that's complex. Where do I deploy? How do I deploy? What does the guarantee look like, and how long is it going to take?" Those are the questions that, really, my team's specifically here to help somebody answer.

(Joel Wineland at 00:42:12) The other side of that coin is, "I have racks, but they're hard to support." A lot of our clients are in lightly manned facilities or facilities where they only have smart hands resources that aren't their actual people, right? Whether you're in Singapore or in Amsterdam or in Salt Lake City, I see a lot of places where people are struggling with, "I don't really have the wherewithal to put resources at every one of these locations and train them on my infrastructure."

(Joel Wineland at 00:42:36) So, "How can I get something that's defensibly supportable in a remote location without a lot of headache and without a lot of risk?" So I think in those scenarios, that's really where we stand out. And I'd say, ultimately, whether it's me or one of my competitors in the space, they should really start to think differently about how they deploy. If everything's buttoned up, if everything's well documented, if everything is standardized and locked into a kind of designated configuration, then ultimately, it'll make that so much easier. You'll be able to get those things lit on the ground like clockwork rather than having to piecemeal it through and have outages that are relatively constant because you can't send smart hands into a facility and have them address an outage with the system without causing a bunch of other problems in a rack when it's not set up appropriately. Right?

(Joel Beasley at 00:43:17) Yeah. Get a little gremlin in there. Racklive.com—is that the website?

(Joel Wineland at 00:43:22) Yes.

(Joel Beasley at 00:43:23) Boom. There we go. So people can go to racklive.com. That's a great name. It's easy to remember.

(Joel Beasley at 00:43:28) And then they can just get little pop-ups. They can request quotes, learn more, or we can put your link to your LinkedIn, things of that nature in the show notes so people have access to it really easily. This is great. Last question here: best leadership advice you've ever received?

(Joel Wineland at 00:43:45) I'm not sure I comply with this one, but I'd say it's still the best is, "Don't try and do it all yourself." So, you know, we're a small team. And while I do have 13 in my group to do various aspects of what we're doing from a production standpoint, and I love to punch keys, I'd say the best advice I've gotten is learn how to delegate.

(Joel Beasley at 00:44:06) I love it. For me, it's like I tend to go there too. So at first, it was like, you know, like a diet. It's unrealistic just to change everything at once. So it's more about catching myself, catching myself in the act and then being like, "Okay, I need to bring in someone else." No, that's good advice.

(Joel Wineland at 00:44:23) You know, I'll give you one other one, actually. So my old manager at Rackspace used to say, "I'd build a battleship to cross a stream." So know the problem and attack the problem, not the idea of the problem, would be the other thing that I'm definitely guilty of on almost a daily basis still. Although it's more fun to build all good things, right?

(Joel Beasley at 00:44:43) So that was his advice to you?

(Joel Wineland at 00:44:46) His advice to me was, "Don't build a battleship to cross a creek," right?

(Joel Beasley at 00:44:49) That's well said. I'm going to write that one down. Thank you so much for listening. And if you found this episode useful, please share it with a friend or colleague who you think would get value from it. And if you have topics that you'd like to hear discussed on the podcast, either add me on LinkedIn or send me an email: [email protected]. Every time I get an email or LinkedIn message, it absolutely makes my day and inspires me to keep going.