How Distributed Storage Disrupted the Enterprise

Can enterprise IT run resilient, cloud-like infrastructure on standard commodity hardware?

In this episode of I/O You an Explanation, host Dan Ciruli (VP, GM Cloud Native) and Steve Carter (Director of Product Marketing) trace the origins of Nutanix AOS from Google File System principles to modern multi-cloud operations. Discover how hyperconverged architecture eliminated single points of failure in IT infrastructure and why Kubernetes is maturing for stateful applications.

CHAPTERS 

00:00 Intro & Tech Flashback 

01:14 How Collaboration Tech Changed the Workplace 

04:08 Early Nutanix History & Employee #33 

05:31 Why Monolithic SANs Created IT Bottlenecks 

11:08 Understanding Hyperconverged Infrastructure (HCI) 

12:54 How Nutanix AOS Mastered VDI Workloads 

15:41 The Full-Circle Journey to Kubernetes & Microservices 

17:28 Hot Take / Cold Take: Kubernetes for Stateful Applications 

RESOURCES

HCI Guide: https://www.nutanix.com/go/what-is-nutanix-hyperconverged-infrastructure-download

Kubernetes Guide: https://www.nutanix.com/go/set-a-course-for-success-with-kubernetes-on-nutanix-download

Watch Next

What is an AI Gateway and Why Do Your LLM Workloads Need One?: https://youtu.be/JED5IZlaR7Y

What Is Model Context Protocol (MCP) and How Does It Work?: https://youtu.be/rMAfBKLZkTk

#Nutanix #CloudNative #Kubernetes

Transcript

Copy Transcript Copied

 00:00:03:15 - 00:00:24:16
Unknown
Hello and welcome to. I owe you an explanation. I'm Dan Ciruli.

00:00:24:18 - 00:00:49:20
Unknown
Hello and welcome to I/O You an Explanation, the podcast where we take on a topic in technology, dive in and give you an explanation about what it's all about. I'm your host, Dan Ciruli. I lead product management for Cloud Native here at Nutanix and today I am lucky enough to be joined by a special guest, Steve Carter, one of the earliest employees from Nutanix. He’s been here a long time

00:00:49:20 - 00:01:15:13
Unknown
and we're going to dive into distributed storage and AOS. What the name is now, what the name was then and why we don't mention it anymore, but really, what it's all about. Welcome, Steve. All right. Thanks, Dan. Good to be here. But before we get started, the way we like to get things going at I/O You an Explanation is by taking on a technology that we all know, something that we use, maybe at work, maybe in our daily lives, and talk about the before times.

00:01:15:14 - 00:01:39:06
Unknown
A technology that came in at a really changed the way you live, the way you work. I've chosen for you, video conferencing. You're not as old as I am, but you've been around longer than video conferencing has been around, or at least in its present form. Talk a little bit about how your life changed from, you know, the days of phone calls and the days of, you know, I don't know, early WebEx to to now.

00:01:39:07 - 00:02:10:20
Unknown
Yeah, absolutely. Yeah. I mean, I guess even, you know, pre-WebEx, it was conference calls, right, where there'd be a number and a pin and you know, you dial-in and everybody would talk over each other. And it didn't work that great, right, unless you really needed it. But I mean, yeah, I think the biggest thing that it enabled is, you know, for me is like working flexibly, I guess you could say, whether that's from home, whether you know, you're on the road, you can still join meetings and still stay connected.

00:02:10:22 - 00:02:29:08
Unknown
You know, in fact, when I was at Salesforce, which is where I was before I joined Nutanix back in 2010, we had one team member who was who was remote, who was in Austin, Texas, and they even set up a little, little monitor in the office and it was like pointed directly at his face. So when he was working, he'd have like a little, Oh, he was like, present? Yeah.

00:02:29:08 - 00:02:51:22
Unknown
So he'd have like a little virtual presence that was like in the next to the, you know, where everybody else's desks were located and whatnot. And so, yeah, so that was kind of my first experience really with, you know, more, I guess, than just video conferencing, but really bring in someone remote into the team. Yeah, do you think it has changed how you're able to collaborate with people who are far, far away

00:02:51:23 - 00:03:11:23
Unknown
compared to what it was like early in your career. Yeah, absolutely. I mean, these days, you know, the folks that I work with are all over the place, right? In fact, you know, I'm in, you know, in product marketing for NKP and the, you know, technical marketing counterpart is over in Amsterdam, right? And we just had a video conference this morning, you know, just like we were right there

00:03:11:23 - 00:03:28:22
Unknown
and that would have been unimaginable, I feel like back, I don't know, 15 years ago or it would have been a lot more difficult, that's for sure. Right, everybody pretty much had to be in the same office. Yeah. Almost all the time. Exactly. You might have, like two separate offices that were sort of centers of mass, but then people would fly in for meetings and that kind of stuff.

00:03:28:23 - 00:03:30:04
Unknown
How about personally?

00:03:30:06 - 00:03:46:23
Unknown
You big FaceTimer? Only during....no...with my parents. They really like to... Do you FaceTime with your parents? Yeah. Yeah, they really like to FaceTime and see, you know, see the family, see the kids, see, you know, It’s a big deal. Yeah, it is. I remember the first time I saw FaceTime, that’s literally what I thought is. This is, this is, this is invented for grandparents.

00:03:46:23 - 00:04:10:04
Unknown
Totally. The idea that grandparents who live far away can see their kids. They don't even want to use the phone now. It's always like just FaceTime incoming. Yeah. So. Yeah. Great. Well, thank you very much. That was a great dive in the before times. Now let's let's come to the present. Actually, we're not going to come to the present because I think in order to explain AOS to us, you're going to have to  give us a little bit of a history lesson.

00:04:10:04 - 00:04:37:10
Unknown
Now, you were early on in Nutanix. I said that before, not just early on. Your employee number 33, a Yep. double digit employee number is extremely low. AOS was I think the first product. It was the first thing. So you were back in the beginning and you remember what it was like. Tell us what what problem they were trying to solve when they...when... when they did this and how AOS was created to solve it.

00:04:37:10 - 00:05:05:11
Unknown
So yeah. So you know, the problem really that they sought to solve was the storage problem was the fact that, you know, storage had become the biggest challenge in IT data centers, not, you know, there was performance challenges, there were cost capacity challenges. But I think the biggest thing was just complexity, right. Like most of the budget and the focus from the IT team ended up going into storage or storage becoming the bottleneck in so many different ways.

00:05:05:13 - 00:05:31:15
Unknown
And so, you know, the founders, one of them in particular was at Google working on the Google File System, and I think, you know, had an epiphany that this technology could be used in the enterprise and could really change the way that enterprises run their systems and really simplify them and make them more, more cloud like. And I spent some time at Google too, GFS, of course, the underpinnings of so much.

00:05:31:16 - 00:05:54:13
Unknown
Yeah. How is that different than what was going on in the enterprise at the time? Yeah. So the way storage worked at the times, the storage area network was really, you know, the technology that had come to dominate IT storage. And it was basically a big monolithic storage array connected over, you know, a pretty complicated fiber channel network is what it's called, right?

00:05:54:13 - 00:06:16:05
Unknown
Which is like a lossless, you know, very high performance network. Yeah, HPC they used hybrid channel. Yeah, yeah. And it's used it's still used today for you know, I look at it actually, you know, maybe jumping forward a little bit kind of like the mainframe where it's not going away. Right. It's still going to be around, you know, probably for a very long time.

00:06:16:07 - 00:06:32:19
Unknown
But it's, I think gone to the point where it's no longer the standard. Right. But that's because, you know, I think HCI has led to a better way to, to build IT Data centers. Okay. So they were...it was dedicated hardware for storage, essentially dedicated networking.

00:06:32:19 - 00:06:39:01
Unknown
You weren’t using fiber channel for and it was a totally separate network, separate networking technologies.

00:06:39:03 - 00:06:42:03
Unknown
And so you have like a whole separate team managing it as well

00:06:42:05 - 00:07:03:15
Unknown
and it created bottlenecks. I think that's the other big thing. So so much of the architecture and kind of engineering focus was figuring out how to scale the performance. You know, given this, this centralized storage array. And as you added more servers and more workloads now, your existing workloads would take a hit performance wise.

00:07:03:15 - 00:07:20:02
Unknown
And all of a sudden it became risky to add or change anything. And what was different about the architecture of...the way...Google File System was built and then later AOS. So the big the big difference is that it's distributed, right, that the storage is located next to the application and so as you have more applications, you're adding more storage.

00:07:20:02 - 00:07:49:17
Unknown
And the way the real special sauce is the algorithm in the back end that keeps the data protected without causing hotspots, without causing certain, you know, sections of the storage, I guess, to become slower, to become bogged down. Ok...So when you say running with the application literally on the same server, it's the same physical servers... Yeah, exactly. Where you're running your applications, the storage is on the disks or flash in those those servers.

00:07:49:18 - 00:08:15:07
Unknown
Yeah, exactly. Okay. So you know, you're no longer using separate hardware and then of course, the networking is taking advantage of the network that's built into those servers as well. Exactly. So it still has to communicate between the different servers. But but now we can just use the standard Ethernet network. You know, it was the timing worked out that ten gig network connectivity was really becoming the standard at that point, right back in the kind of one gig era, it probably wouldn't worked as well, right?

00:08:15:07 - 00:08:35:18
Unknown
And now we're up to 100 Gig, you know, plus and it's become just, you know, a non-issue. But yeah, but the ability to...10 Gig is consumer now. Yeah exactly. People have that in their houses. Right? Right. I have a friend in Willow Glen who has ten gig actually to his house. I have 10 Gig to my house. I only have a 1 Gig network in my house. That's amazing.

00:08:35:21 - 00:08:55:17
Unknown
Yeah. In theory I could upgrade some prosumer stuff. There you go. Yeah, but, yeah, I mean, so I think what that did was it made it easy to. It made it easy. That's the big thing to to manage the infrastructure so that IT teams can focus then on adding value on the applications themselves and less on, you know, troubleshooting and, and so on.

00:08:55:18 - 00:09:17:14
Unknown
And then the servers that it was running on Linux boxes. Yeah, just standard commodity servers. Right. That was the whole idea that instead of needing to go, you know, bu y a, you know and that was I think one of the big things with storage arrays is that they had had very customized hardware. It was designed to be really durable to, you know, never fail, right.

00:09:17:14 - 00:09:38:05
Unknown
And so on, which, it still fail but, you know, the idea was to to minimize the chances of a failing. And then you only had essentially two typically in the standard architecture where if one failed, now you've you only have one left and you better hope that it doesn't fail, right. The difference with this architecture is that it's duplicated across every server.

00:09:38:05 - 00:10:02:14
Unknown
So if a server goes down, you have 20 other servers that are still there to pick up the load. And so it makes it so that you don't need that super durable hardware anymore. The hardware can fail and the system will continue to operate without anybody noticing. Right. And then I remember I wasn't a Google yet in the early 2000s, but I remember very clearly people talking about, you know, one of the amazing things was that they had built this really resilient, really durable software on commodity hardware,

00:10:02:14 - 00:10:23:18
Unknown
when it's time people didn't do that, people turned. I you know, I also spent some time in EMC, right. And they used to be a joke in Silicon Valley when you got your your first VC check, you then went and just wrote checks to EMC, you know, to Cisco and to maybe HP at the time for your things and, and Google said, no, we're just going to build this all out.

00:10:23:20 - 00:10:46:12
Unknown
Right, right, right. Yeah. So so they they started Nutanix. They wrote AOS, which is not you know, it's a separate implementation from Google File System. Yeah. Yeah. And what did they call it? So that's a good question. I don't even remember what the first name was. I remember that for a while. It was NOS, NOS, Nutanix Operating System

00:10:46:12 - 00:11:18:16
Unknown
and it had the kind of like Nitrous Oxide visual. You know, it was trying to evoke this high performance thing. And then and then Acropolis was actually maybe that's one of the terms that a lot of people think of as, you know, you know, the Nutanix product, but Acropolis is really the hypervisor actually. So but then that then, you know, kind of long story short, was was rebranded into the core platform as Acropolis Operating System, which then we've completely gone away with.

00:11:18:16 - 00:11:52:22
Unknown
and now it's just AOS. And you said before HCI, for people who are listening at home and don't know this term, HCI, Hyper Converged Infrastructure, right. Did that term exist before Nutanix or did it...did it happen at Nutanix? You know, what I recall happening was that it was an industry term that maybe was coined by Gartner or one of the analyst firms talking about Nutanix and the other competitors in our space that were kind of, I would say, the fast followers of us, right, that were saw what we were doing and saw that there was a bunch of potential there.

00:11:53:00 - 00:12:21:21
Unknown
And then we latched on to it. So, you know, my recollection was it wasn't somebody in Nutanix that came up with that term. But, you know, we we latched on to it pretty quickly because, you know, and what it came out of was converged infrastructure, which was the big, you know, movement in the industry before us. VCE, right, with VMware, Cisco and EMC partnered to say, we're going to take our storage array, hypervisor and servers and package them into this SKU that you can buy all in one.

00:12:21:21 - 00:12:39:02
Unknown
And this should make it easier to manage, make it easier to buy and all that. It turns out it actually made it more complicated to buy and manage, because if you wanted something customized, they'd have to go back and validate everything and it would delay things by months. So it turns out converge infrastructure did not, you know, meet the promises.

00:12:39:04 - 00:12:59:05
Unknown
And so then this idea that taking it another step further and hyper converging, bringing them actually together in the same box was kind of the the evolution of... So one set of hardware  but it's running your virtualization. It's running your your storage and it's running your, your network. Yeah. All on that same hardware. And by the way, you didn't get into it, but it's ethernet, is the network right?

00:12:59:06 - 00:13:31:05
Unknown
Yeah. It's the standard Ethernet. Yeah. Exactly. Same same network that's already there. Same network that’s already built in those servers. And I heard that while it wasn't necessarily designed for it, the first killer use case was virtual desktop infrastructure. Is that is that true? That is true. Yeah. And one of the things, actually, that I really loved about being part of Nutanix in the early days was, was how I think originally there was this vision that we were going to take over the IT data center, that every enterprise was going to be hyperconverged in the future.

00:13:31:06 - 00:13:53:11
Unknown
Right? And that you'd have this distributed Google kind of across the globe. But we had to start somewhere, and there was this use case that was decidedly, you know, not that exciting, right? Right. It's like like desktops being delivered over the network so that you didn't have to put a physical, you know, actual physical desktop on each person's desk.

00:13:53:13 - 00:14:11:16
Unknown
But the problem was that the existing technology did not work well for it. So that was, I think, one of the reasons why I worked so perfectly was because the SAN was just not designed to handle, you know, hundreds of desktops, all logging in and booting up at one time. Everyone comes on and turns it on. Right. And now you talk about bottlenecks before like that is hitting that bottleneck.

00:14:11:17 - 00:14:36:04
Unknown
Exactly. So the boot storm is what is what they call that. And boot storms used to be the biggest challenge in VDI. And I don't really think you hear about boot storms anymore, right? Literally, I've never heard the phrase. Yeah. And, you know, the big reason is because if you have, again, let's say 500 VDI desktops and it spread across ten nodes, you know, now you just have, you know, whatever it is, 50.

00:14:36:05 - 00:14:57:16
Unknown
Did I do the math right? 50 desktops per node. Right. Versus...and they're not all hitting the same storage device. Yeah. So they're all just hitting that local storage controller versus all 500 going at the same, you know, the same storage control. So they hit local storage. But there is resilience. There is duplication automatically I assume so that if a node should fail that data is preserved.

00:14:57:17 - 00:15:12:08
Unknown
Yeah, exactly. And that's where it comes to like the difference of reads versus writes. So when you're reading data you can just read it locally. That doesn't impact anybody else. Right. And that's really one of the biggest benefits of storing the data locally is you don't have to go across the network for reads and writes and then writes

00:15:12:09 - 00:15:31:01
Unknown
of course, do need to be duplicated. And it has a really cool algorithm with the way that it actually, in real time, decides where to send the replica copy based on the load of the other nodes in the cluster. So if one node has a bunch of desktops booting up at once, the system will decide to send the replica to other nodes that are less loaded.

00:15:31:01 - 00:15:57:05
Unknown
Very cool. Yeah. Very cool. All right, so you've been here. What year did you start? So 2011. Okay. We're in now in 2026. Yeah. It's crazy. 15th anniversary coming this year. That's right. I think you get a gold watch or something. Well, I would if I hadn't left for six months, I suppose. Probably so. We all make mistakes. And what's the coolest addition you've seen in the in those years?

00:15:57:05 - 00:16:26:15
Unknown
So started with AOS still really I think kind of the core. What cool has happened lately? Well, lately, I mean my passion now is certainly Kubernetes and NKP, right? And I don't say that just because I'm talking to you, but but truly, I think I mean, in fact, I remember in the very early days when, you know, we were all working at the like when I, when I joined, by the way, they gave us a laptop with no hard drive and also and they basically like, here's your hard drive, go install it.

00:16:26:15 - 00:16:51:17
Unknown
And then here's an ubuntu CD like, you know, hopefully you'll be done with that by lunch and...you joined ands started an SE, so everyone knows. Actually, no, I was QA at the time. You were QA first. So I was yeah. Wow, you’ve done a lot! And I was working on the test automation framework. When we bring you back we’re just going to talk about your career. There we go. Yeah sounds good but and so so so you know it was like that early startup buzz we're all excited about the future and whatnot.

00:16:51:17 - 00:17:10:14
Unknown
And I remember people talking about containers and how this new container architecture I remember that Mesosphere actually and Docker with a big, you know, the hot topics at the time and how this architecture would lend itself perfectly to that. Now, we were way too early to start getting involved, but it's funny how it's it's really come full circle.

00:17:10:14 - 00:17:31:12
Unknown
And we've gone with like, you know, that's I feel like in a lot of ways what Google created that allowed them to create their search engine that scaled so well originally. And now Nutanix had created this kind of storage version of of that infrastructure for the enterprise. And then Kubernetes is then the compute side of that and so now you have that full stack, right.

00:17:31:14 - 00:17:52:10
Unknown
And now the world has Kubernetes. Yeah, exactly. And we used to always go pitch ourselves as Google for the enterprise and at a certain point we had to stop doing that because everyone was like, look, we're not Google. We don't run at their scale. That's too complex. But yeah, yeah, yeah. All right. Well, this fantastic conversation before I let you go

00:17:52:11 - 00:18:10:17
Unknown
like to close episodes with a little thing we call What's Your Take? And I'm going to give you a topic. Give me a take. Is it a hot take. Is it a cold take, a lukewarm take. The take I want from you is Kubernetes is best for stateless applications. Oh, that is a cold take. Tell me why.

00:18:10:19 - 00:18:30:16
Unknown
Because. So Kubernetes containers themselves are stateless and they're designed to be able to fail and then another container will pick up the load, right? But every application has state of some sort. Right. Even...I just think of even, you know, like minesweeper, like if you need to score the high scores that state that has to be stored in a database somewhere.

00:18:30:16 - 00:19:04:06
Unknown
So I feel like it's a little bit of a misconception that even though containers themselves are stateless, the actual app isn't. And there's another point actually I want to make really quick is on AOS. This is one of the things that I learned when we join Nutanix is the way that it was designed is actually very similar to how a microservice is designed with this idea that any time there's an exception or a fault, it kills the process and restarts and another, you know, another version of that service will then take over, rather than trying to handle every single exception and recover without failing.

00:19:04:07 - 00:19:24:04
Unknown
Right. And so the fact that the storage layer itself is actually built in the very same way that the applications are being built in Kubernetes, I think points to the, you know, alignment, I guess, of the technologies. All right, I agree. Every application you don't have an app if you don't have data. Every application is backed up by data and eventually everything needs some state.

00:19:24:05 - 00:19:26:13
Unknown
Kubernetes has gotten really mature at handling that as

00:19:26:19 - 00:19:42:11
Unknown
well. tThanks a lot Steve. Great conversation. Really appreciate you taking time out of your day to come spend some time. And I'd like to thank those of you who have listened to this episode. Really appreciate you being here. We will be publishing these regularly. They will go out everywhere you get your podcasts.

00:19:42:11 - 00:19:50:09
Unknown
And if you want to find them, you'll always be able to find them in Nutanix Newsroom, which is another great place to find more news about our products and what we're doing.

00:19:50:09 - 00:22:37:06
Unknown
So we'll see you soon, and we'll deliver more explanations that we owe you about the technology that really affects the way we work.