KubeCon + CloudNativeCon EU 2026 | Phil Trickovic, Tintri

Clips
More from KubeCon + CloudNativeCon EU 2026

Phil Trickovic

SVP

Tintri

play_circle_outline Tintri: workload-aware storage appliance using ML to tune resources in real time

play_circle_outline Auto-tiering: real-time per-workload tier decisions, not traditional HSM hot/warm/cold

play_circle_outline Tiered NVMe, SSD, and Spinning Disk Strategies with Predictive Capacity Planning for Cost-Effective AI Datasets

play_circle_outline Mission-Critical Edge Intelligence: VMs, CXOS, RISC‑V PixForce Vision and Resilient Local Processing

Info
Transcript

Phil Trickovic, Tintri

Phil Trickovic

SVP Tintri

In this interview from KubeCon + CloudNativeCon EU in Amsterdam, Phil Trickovic, senior vice president at Tintri, joins theCUBE's Rebecca Knight and Paul Nashawaty to discuss how workload-aware intelligence is evolving from the data center to the edge. Trickovic traces Tintri's roots to 2011, when it pioneered what is now called AI-driven storage by building a platform that continuously monitors workload behavior and tunes resources in real time without impacting adjacent workloads. He details the latest auto-tiering capability, which moves data dynamically b... Read more

explore Keep Exploring

How long has Tintri been on the market, and what is its primary value proposition or distinguishing feature? add

How does auto-tiering work—including its feature functionality, failure handling, and optimization—and how does it differ from traditional hierarchical storage management (HSM), particularly regarding real-time decision-making, impact on adjacent workloads, and what’s new compared with earlier SSD-to-spinning-disk tiering? add

How does the new storage-tiering and optimization solution work��specifically its NVMe-to-SSD tiering, handling of AI-generated data sets, and automated x86/network resource planning based on workload history and forecasts? add

How should edge computing and modular infrastructure be designed to support AI workloads—covering differentiation from traditional HCI (CPU vs GPU), pushing intelligence and autonomy to edge devices, model deployment and incremental updates, and ensuring data integrity and continuity when devices lose connectivity? add

bolt Powered by CUBE AI

Phil Trickovic, Tintri

search

Rebecca Knight

>> Good afternoon, everyone, and welcome back to theCUBE's live coverage of KubeCon + CloudNativeCon EU. I'm your host, Rebecca Knight, sitting alongside Paul Nashawaty, principal analyst here at theCUBE. We're live in Amsterdam for day two of KubeCon.

Paul Nashawaty

>> We are. It's super exciting to be here. It's really been an exciting show. 13,000 people around us, it's really exciting.

Rebecca Knight

>> Indeed. Indeed. Well, our next guest has also a lot of exciting things going on at his company. I'd like to welcome Phil Trickovic, SVP at Tintri. Thank you so much for coming on the show.

Phil Trickovic

>> Thanks for having me, guys.

Rebecca Knight

>> So, you've got a lot of things going on. You're juggling some pretty significant initiatives, auto-tiering, new modular compute stack, a push into edge. We're going to talk about all this. But for our viewers who are less familiar with Tintri, just tell us a little bit about your company and what you're all about.

Phil Trickovic

>> Okay. So, Tintri's been on market since 2011. We were the first AI device appliance on market before this was called AI. It was when it was called machine learning. What our claim to fame and our real value to the market was is that we're workload aware. Meaning, that we know what a workload's doing. We know what it's done over time. We gather those historic parameters and we will tune in real time your resources required to run that workload. So, we are the most efficient way to serve ones and zeros.

Paul Nashawaty

>> Yeah. Phil, we've talked a lot in the briefings. We've had a lot of discussions outside of this, so I'm glad to have you on.

Phil Trickovic

>> Thanks.

Paul Nashawaty

>> Let's start jumping into some of the announcements. So, when we look at auto-tiering, there's this misconception that auto-tiering is that just HSM, hierarchical storage management, and this has nothing to do with it. It's hot to warm to cold, right? So, tell us a little bit about the feature functionality, the failure handling, the optimization, because it's very much mature compared to some of these old solutions.

Phil Trickovic

>> It's very mature. And it's not a HSM-type thing where... You know the story with that. It's live. It's real time. It's decided upon in real time. So, as workloads become hot or they demand additional resources, it'll auto-scale the resources available to that X-specific workload without impacting adjacent workloads. That's the trick with the HSM strategy, right? It's not aware of what X workload next to it is consuming. So, we've been aware of that since day one. That's what our special sauce is.

Paul Nashawaty

>> Yeah. Yeah. No, I think it's cool. So, this is new, it's a new functionality that you're putting out.

Phil Trickovic

>> Right. So, before it was just tiering between SSD and spinning disk, right? This was when we launched in '11. Now, we're bringing it back where it's going to go between NVMe, high cost, high performance, and SSD tier, where you may not want that workload to ever be put down to spinning disk or a lower-quality medium. Then, where you've got evolution of this magnificence that's happening with this with AI, the amount of data that these things generate is being overlooked, like massively overlooked. We can clear that up for people and make sure that those data sets that you're going to need for additional trainings or whatever are kept on the most cost-effective media period, as well as enhance your x86 compute stack. I think you and I spoke of this before, dating myself. Started in my journey somewhat in the mainframe world with job schedulers. And when this came out eight, nine years ago, I looked and went, "Wow, this is just a job scheduler for mainframe." But it's what's most needed in x86 because I don't think anybody would argue that we don't optimize x86 at all. So, this allows you to optimize automatically based on what your workload's doing. Not what a sales guy or someone's telling you you need, it analyzes what you're doing, keeps the historics. Then, it will predict what's needed for future use. And that's where you can come in and go, "I want to grow my business 7%," or, "I want to grow my workload farm 2%," hit enter. It'll tell you exactly what you need from networking x86 stack and storage stack on the back-end.

Rebecca Knight

>> And that's going to translate into real savings for organizations?

Phil Trickovic

>> Exactly. Absolutely. So, you're not buying blind and you're not buying based on what some analyst, no offense, or me or anybody said. It's based on your data. It boggles my mind sometimes with Tintri that this hasn't been more...

Rebecca Knight

>> Embraced.

Phil Trickovic

>> Embraced or acknowledged even. You're just throwing money away the old way and it's not the most efficient way. And you're costing our plant. I mean, we're standing up nuclear power plants on beaches now to power searches. Come on, world, let's use the resources wisely or we're never going to get where this can go.

Rebecca Knight

>> I like the way you talk.

Phil Trickovic

>> I'm passionate about it because I see it every day and just go, "Well, someone's burning oil for this that doesn't need to be burned."

Rebecca Knight

>> Right, right. We have the data that shows what you need to do.

Phil Trickovic

>> We have the technology.

Rebecca Knight

>> Okay. That makes a lot of sense. Okay. So, let's move on to the compute stack because you're building something modular here, so talk a little bit about that.

Phil Trickovic

>> Very excited. Over the last 18 months, we've been working very closely with Platform9. I believe you guys know them. Amazing technology, amazing company and amazing people who really want to see this fly and for the benefit of the world, to bring these things to market without consuming every drop of oil that we have. So, we partnered with them about 18 months ago on developing an integrated stack. So, we're trying to avoid HCI, but unfortunately, I don't know what else we can call it yet. We're bringing HCI stack to market with both Intel AMD chip sets, bolted, validated, certified, and supported by both companies jointly.

Paul Nashawaty

>> Okay, let's double click down on this because the modular compute stack, it's about the architecture, the scalability, the performance and reliability. So, when we look at alternatives in the market, with workflows, AI workflows happening, the explosion of applications that are coming out in the market, it's really just driving the need to have that compute on demand, so to speak, and having the ability to fluctuate as you need to. And on top of that, there's also the market shifts of what's happening with your heritage environments. Say you maybe have a heritage VM, or you want to move to more containerization, and you need to have that specific workload to do that. How do you look at that from a modular stack?

Phil Trickovic

>> To us, they're ones and zeros, and how do we most efficiently serve them? So, not to be a smart alec, but that really is, we can do both in the same environment. We're not restricted by you have to do your VMware here or your Platform9 there, and then your Citrix VDI... No, we do it all in frame. We're aware of those workload profiles and adjust their requirements in real time. So, it's moving into the world of AGI, a little bit, .

Rebecca Knight

>> But it's very complex. And so, you need to get the teams the flexibility, but also make sure it's stable too at the same time.

Phil Trickovic

>> And enterprise-class mission-critical. Our customers, we've got mission-critical ones. Many governments, without going into details, but we have many mission-critical environments that we've been serving for many, many years. And yes, I'm greatly oversimplifying how these things work, but this is where we have to go. We won't get to realize what we can and the magic of what's happening here if we don't figure this out. It's just you do the math. It's not me saying it. It's not going to happen.

Paul Nashawaty

>> No, you're right. You're right. There's a lot of things happening, a lot of growth with applications we're seeing in our data. We see 500 to 1,000 applications happening at the edge alone, so we definitely see the need. But let's talk about the differentiations, because when we look at HCI, it's not new, right? It's something that's been around for a long time. But modular computing is something that is more important today than it probably has been in some time because when you think of these AI workloads, they're consuming a lot of resources, right? And these GPU appliances that are going out that maybe you don't need a GPU, maybe you can use a CPU. What's the differentiation that we're looking at?

Phil Trickovic

>> Thanks, Paul. You're leading right into where it's going there. And this is something we're seeing in the market and it's a bit of a vision, but to push to the edge, right? To us, our long-term goal is the server, what we call servers layer now is going to not be a server layer anymore. These devices can be pushed out. Another announcement or partnership, I don't know if we talked about with PixForce. We've been working with them for two years on camera vision, basically, and sensors and a bunch of other things with it, but again, ones and zeros. We've made Risk five based devices. We bring up RISC-V, you guys are familiar with... Okay. These are going to be auto deciding on the edge for some companies which pushes us into an AGI-type world where they're making a decision in real time. I don't want to be hyperbolic with this. Yes, it's going to be two or three things, but it's really autonomous and it's going, "Hey, Paul doesn't have his glasses on today. Could his glasses be there? Should we alert him now that he's got the wrong color?" They're little things, but they're very important for industrial use cases. It's predictive.

Paul Nashawaty

>> Predictive?

Phil Trickovic

>> Predictive, yes. And we're seeing that and we're starting to deliver it, but that's happening on the edge. It's not happening in the middle of the cloud or the data center. So, we've extended CXOS out to support those devices so that we can pull those intelligence consistently and historically and retrieve them and then adjust in real time. So, that's the next step of where we're going with all of this and where we see it ending up. It's going to be very interesting how devices are consumed.

Rebecca Knight

>> But devices, as you point out, are often on the move. Vehicles are a very good example. So, how do you make sure if they, for instance, lose connectivity, that the data doesn't fall through the cracks?

Phil Trickovic

>> Stays with the device. I mean, I can whiteboard it for you, but there's always those things. And yes, of course, as we do this and you're seeing it more from an old guy who-

Rebecca Knight

>> You're not that old, Phil.

Phil Trickovic

>> Well, I've been doing this a while, but when RAID was, "Absolutely can't have a system without RAID," now you're seeing all these new bare metal things come out and those RAID 0. So, some of the requirements still need to shake out, but the way we've architected, you're always protected in three or four locations, period. Now, if you had a PX4 smart cam in your car and you lost your... Not SpaceX. Man, what's the-

Rebecca Knight

>> Satellite OQ.

Phil Trickovic

>> Satellite. Yeah, you lose it for 30 seconds. Whatever we capture there, we're going to have to catch up to, but it's possible to do it. The bandwidth now is not a constraint to keep those up in real time.

Rebecca Knight

>> Those small connectivity drops?

Phil Trickovic

>> Right.

Rebecca Knight

>> Yeah, okay.

Phil Trickovic

>> And remember on these devices, what we're seeing now is they don't have to come back. They just report a flag status. So, they're deciding, they're our brain. And while they're only looking for three or four things to report on, they're not having to constantly go back and retrain the model. I don't know if we want to go into that whole talk, but the way we're doing it now with training these big models, once you've moved them out to the edge, they just need minor incrementals in some cases. If we're smart on what it's actually doing, if we're selling this that's doing everything and it's got every LLM on it, I mean, good luck, but you're burning a lot of power.

Paul Nashawaty

>> Yeah. To Rebecca's point, I think that there's a lot there to unpack. If you're looking at local processing and centralized processing, that requires different types of resources. So, you have different T-shirt sizes, right? So, if you're at the edge... I was talking about the number of growth and applications, but the reality of it is the skill gap at the edge is you don't have a full tech team at the edge, right? So, you need to deploy by dropshipping something, plugging it in, making it work, and you line a business at the edge, whether it's retail, finance, insurance, whatever it may be, they're not technical people, but they're writing code. They're builders, right? And those builders need that processing power to process compute at that level. So, you can add in the storage and the compute and everything that needed at the appropriate level. That, I think, is something that's interesting from the Tintri conversation that we were talking about, I think it's very interesting.

Phil Trickovic

>> Yeah, because we've extended all of our optics and our functions and abilities out to those devices. So, it is a little complicated, I suppose, to explain here, but the key to that is understanding what that device is supposed to do and what is it sending back.

Paul Nashawaty

>> Yeah. Yeah. And I think it's important because of the amount of data that's being processed at the edge, not everything can be sent back to a core location-

Phil Trickovic

>> Or in time to do it.

Paul Nashawaty

>> Or time.

Phil Trickovic

>> Yeah, to be impactful. How do I explain it without using a company name? Oil and gas, for example. Were you with us at OTC at Houston?

Paul Nashawaty

>> Yeah.

Phil Trickovic

>> Okay. So, at OTC, you're seeing... It's really amazing. I can't use the name, but in the Gulf of Mexico and also off the coast of Brazil, we've got a customer that has brought on basically per people-less oil rigs, but the amount of sensors that are required to stay in compliant is really, really insane. However, it works. And we've got these smart cams with pics out there that are sensing methane in the water, in the air, and I don't know the exact amount, but it's 0.001% per million. What's interesting is when that changes. So, when that changes, what do you do? So, now that's where the AGI starts to come into play that it already knows that if that goes above this level, do this, that level, do that. And this extrapolates out to millions of different functions, but they're all happening on the device. They're not going back to be retrained. So, what we do after that is that change state, we call it, gets sent back, just the change state, not everything that happened. Just, okay, this parameter was hit. We have to adjust the model for that when we're ready to. But as far as the functionality of not letting methane escape from this, it's never interrupted.

Rebecca Knight

>> So, Phil, stepping back and looking across all three, the auto-tiering, modular compute and edge computing, where do you think Tintri is furthest along and what are the problems that are still the hardest to solve? Or does it just go back to the sustainability issues that you've alluded to?

Phil Trickovic

>> That's a really good question. Obviously, we're very strong in cloud. The basic fundamentals we've got, it's table stakes for us at this point, right? Where we're seeing the big push and extensibility happening is out to those edge types devices because these were served out of huge data centers before with big latencies across the... That can be eliminated and there's a lot of cost that goes out the window when you do that because these devices are not expensive and they have GPU cores in it, just enough to do what they're supposed to do. And that's what we're seeing everything moving to and that's where we're going. Still requires a very large cloud component, still requires an on prem component. VMs aren't going anywhere. It's like 390, whatever it is now, it's still... Credit cards are still going through mainframes, ladies and gents. VMs aren't going anywhere. It's going to grow, but this particular piece is just fascinating because that's really going to drive the rest of the innovation behind it.

Paul Nashawaty

>> All right. I want to ask a question, but I want to tee it up for our next conversation because I think that the question, since we're in Europe, and we can't have a conversation without talking about data sovereignty-

Phil Trickovic

>> Absolutely....

Paul Nashawaty

>> governance, compliance, and regulations. We're in Europe, so we have to talk about that. So, in a short sentence or so, and we could tee it up for the next segment, what is your response for data sovereignty and those compliance issues?

Phil Trickovic

>> Absolutely needed, and it's not a restriction as far as we're concerned on any of this that we've just discussed. Fully understand it. We've been dealing with this also for the last 10 years. Not even just GDPR. India, you have BIS. I mean, every country. Japan has its own set of challenges. That's not an issue for us right now. And we're encouraging it because this free-for-all of absorbing everybody's IP isn't going to sustain either. So, no, we're 100% behind that, but we can do either way. We can do private, public, hybrid. Table stakes.

Rebecca Knight

>> Well, that is a terrific segue because we are next going to be speaking to you again about Tintri's hybrid cloud-enabled storage. So, I look forward to that.

Phil Trickovic

>> Me too.

Rebecca Knight

>> Stay tuned for more of theCUBE's live coverage of KubeCon + CloudNativeCon EU. I'm Rebecca night for Paul Nashawaty. You're watching theCUBE, the leader in enterprise tech news and analysis.