KubeCon + CloudNativeCon EU 2025 | Andrew Hainsworth, Aussie Broadband & Luke Gentele, Loft

Clips
More from KubeCon + CloudNativeCon EU 2025

Lukas Gentele

Co-Founder & CEO

LoftLabs

Andrew Hainsworth

Principal Platform Engineer

Aussie Broadband

play_circle_outline Loft Labs virtualizes Kubernetes and cloud native stack for maximizing hardware efficiency.

play_circle_outline Aussie Broadband using vCluster for efficient multi-tenancy and resource utilization.

play_circle_outline vCluster reduces CPU cycles and VM usage, saving significant resources.

play_circle_outline vNode technology provides deep isolation and security in multi-tenancy environments.

Info
Transcript

Andrew Hainsworth, Aussie Broadband & Luke Gentele, Loft

Lukas Gentele

Co-Founder & CEO LoftLabs

Andrew Hainsworth

Principal Platform Engineer Aussie Broadband

Exploring virtualization in Kubernetes, this video features insights from KubeCon+CloudNativeCon EU 2025. Lukas Gentele, co-founder and Chief Executive Officer of Loft Labs, joins alongside Andrew Hainsworth, principal platform engineer at Aussie Broadband, at this event hosted by theCUBE. The discussion delves into the complexities and advancements in Kubernetes and cloud native technologies.

In the video, hosts from theCUBE, including Rob Strechay, engage Lukas Gentele in a discussion about Loft Labs' work in virtualizing Kubernetes clusters. This ap... Read more

explore Keep Exploring

What features are emphasized in virtualizing Kubernetes and the cloud native stack? add

What solutions were considered for effectively partitioning resources in a multi-tenant environment on bare metal infrastructure before ultimately deciding on vCluster? add

What were the resource cost implications of deploying the equivalent number of CAPES clusters using VMs compared to other options, specifically in terms of CPU cycles and number of VMs saved? add

What are some challenges and concerns with multi-tenancy in Kubernetes clusters, and how is the issue of container breakouts being addressed in a more efficient way without the use of virtual machines? add

bolt Powered by CUBE AI

Andrew Hainsworth, Aussie Broadband & Luke Gentele, Loft

search

Rob Strechay

>> Hello, and welcome back to KubeCon CloudNativeCon, London 2025, Rob Strechay. We're here getting through day three of our coverage wall-to-wall around everything that's been going on in the developer lifecycle to help people understand, how do you make your life easier with all of the different projects and companies that are here exhibiting at KubeCon? And I'm really excited, I have Lukas Gentele. Did I say that properly?

Lukas Gentele

>> That's perfect.

Rob Strechay

>> Awesome. Lukas Gentele, who's the co-founder and CEO of Loft Labs, welcome to the show. And Andrew Hainsworth, principal platform engineer, Aussie Broadband all the way over from Brisbane so welcome onto the show here.

Andrew Hainsworth

>> Thanks for having me.

Rob Strechay

>> So this is really cool because again, I dug into your stuff a little bit, really interesting things. But for those at home who don't know who Loft Labs are, break down what you are and what are the different products that you bring to market?

Lukas Gentele

>> Yeah, we're all about virtualizing Kubernetes and the cloud native stack. And with virtualizing, I don't mean VMs. I actually mean the opposite. Bare metal Kubernetes, for example, is a big case for us. What we allow you to do is spin up virtual Kubernetes clusters on top of traditional Kubernetes clusters, right? So kind of like a VM, but the Kubernetes way.

Rob Strechay

>> So it's really helping people maximize their hardware and their investment in that so that. Because I mean, right now we see it budgets are only going to grow worldwide by 3.5% this year. That's down from 4.5% that they did last year. So I think again, when you start to look at this, it's really getting more out of that and doing it in a secure and safe way. Again, bring you in here, Andrew. You're using Loft Labs. You're using vCluster. Why? What has led you on this journey with them?

Andrew Hainsworth

>> And what a journey it's been. We embarked on out of a container story probably only two and a half years ago. I came into a team that was purely VMware, top to bottom compute storage network, very, very traditional. And we decided to go with CAPES as an orchestrator. Started rolling it out on our bare metal infrastructure, converting VMware hypervisors to bare metal CAPES. Subsequently, we had some challenges with multi-tenancy and how do we effectively partition these resources where we have a finite amount of bare metal? So my team and I, mostly my team, evaluated a number of ways of doing it.

Andrew Hainsworth

Namespace segregation with RBAC, which is quite clunky and clusters on VMs, which is resource intensive. All those kernel copies, other tools like Kiosk and later looked into Comagi as well. But suddenly we came onto vCluster and it just clicked the way it does what it does, and the ease of deployment and the self-service options that came with it for our customers. It was just the way to go.

Rob Strechay

>> You also had some impressive results from this because that to me was one of the things that stuck out when we caught up a little bit before. How much more efficient is this for you than your previous?

Andrew Hainsworth

>> It's tremendous. We did a comparison recently on what would this cost us resource-wise to deploy the equivalent number of CAPES clusters using VMs? Because we were looking at that from the perspective of what would VMware licensing look like compared to these other options? And it was somewhere in the order of 50% to 60% fewer sort of CPU cycles to achieve the same workloads, 100s of VMs saved. We're looking at 40 vClusters that we're running now, and they have a variable number of nodes, but say five, six, seven per vCluster. 100s of VMs and terabytes of memory really that were saved.

Rob Strechay

>> So Lukas help us understand what a typical customer looks like when they engage with you guys and why they engage with you and things of that nature so that you can really just people understand the use case, I guess you could say.

Lukas Gentele

>> Yeah, I think we have to differentiate between the public cloud and the private cloud, right? In the public cloud, it's really about how can I spin up clusters without spinning up more EKS or AKS clusters, right? They're all very expensive. There's a lot of duplication, right? You run all these tools like OPA logging, monitoring, Prometheus. Everything is duplicated in each cluster, right, so it's not just the cluster feed. It's also underutilization of nodes, right, underutilization of things like Istio, OPA, all of these things, right? Consolidating things into fewer clusters and then having that virtual cluster layer on top that still allows you to do multi-tenancies, split things up into separate units is incredibly valuable. I think in the private cloud, the challenge is more, okay, we have all of these virtual machines. They consume a lot of CPU and memory.

Lukas Gentele

There's a license fee attached to it that is just going up into the right, as we all know, right, recently and customers just want to eliminate VMs, right? They want to go bare metal. They want to create one bare metal Kubernetes cluster ideally. Maybe it's multiple data centers, so maybe you have five Kubernetes clusters, but you don't need 200, right? And then you want to spin up these virtual clusters dynamically on top. I think that's been a driver for a lot of folks. And then last but not least, CPU clouds, GPU clouds, right? Not just CPU cases, but also GPUs is a really big driver because you don't want virtualization when you're working with GPUs, right? You want the bare metal performance. There's about a 30%, 40% overhead latency when you're using separate kernels virtualization and getting rid of that with GPUs, these pieces of hardware is so expensive. You want to get every little bit of efficiency out of them. That's what we can help you do.

Rob Strechay

>> Is that some of the direction that Aussie Broadbrand is going in? You're looking at this and saying, "Hey, this is how we get to the future of our infrastructure." Is that something you're seeing?

Andrew Hainsworth

>> Do you mean in terms of-

Rob Strechay

>> Yeah, like you're ....

Andrew Hainsworth

>> adding GPUs and?

Rob Strechay

>> Yeah, preparing for AI and AI use cases and stuff like that.

Andrew Hainsworth

>> Well, obviously, it's a big topic at the moment, and I think if you talk to some of the people in our organization that they certainly have designs towards getting some of that, our own sort of inferencing and potentially LLMs on-prem to deal with the data that we have ourself. I wouldn't say it's a absolutely primary consideration for the platform we're implementing now, but definitely it is in our minds that it might be something we'd like to do.

Rob Strechay

>> Yeah, and you're not alone, by the way. I did a panel for the data on Kubernetes group on what day was it? Wednesday. The days are all floating through my head here. And we asked the 600 people who were in the room, how many were actually doing training versus inference? And it was 10 people were doing training or 10 organizations were doing training, and about 15 were doing inference.

Andrew Hainsworth

>> .

Rob Strechay

>> So I think what we see also in our data is a lot of people are experimenting with this, and one of the big places they're doing it is on things like a vCluster so that they can get efficiency and get standardization. Is that some of the stuff that you're also seeing in your customer base?

Lukas Gentele

>> No, absolutely. I think if people are not looking into that trend and considering the use of GPUs and AI and building that kind of capability out, yeah, I think then they'll probably be behind in a couple of years so I think everybody's looking into it. It's obviously early days still, right? And I think a lot of deployments may not be as efficient as they should be, but hopefully we can make a contribution. And especially with our KubeCon announcement here with vNode, I think that's another big contribution towards making GPU nodes and CPU nodes more efficient as well.

Rob Strechay

>> Yeah. Let's dig into that because again, that was actually where I was going to go next is help people understand what that is and why it's important? And why things like kernel level and node level virtualization are really important for this type of thing.

Lukas Gentele

>> Yeah. When you're thinking about multi-tenancy, right, we virtualized the Kubernetes control plane so you can spin up these Kubernetes clusters inside containers, right, but you're still going to have multiple Kubernetes clusters on the same underlying traditional cluster. And these virtual clusters are going to schedule inevitably pods on the same node, right? So you have shared nodes, and if you are security conscious or you want to prevent noisy neighbor problems, you want to shield against container breakouts, right, those are very pressing problems in multi-tenant environments. And so far, we've only been at the Kubernetes API layer with vCluster. Now we're going a level deeper and help you do that multi-tenancy in the right way without virtual machines, right? There's these approaches like cutter container VMs and VMs or micro VMs, right, these kinds of approaches. There's gVisor, which is second filtering, and all of them have their downsides and they have a lot of loss of efficiency, performance issues, latency, right? So we decided to build our own technology, because we've seen our customers struggle with these technologies, and we want to bring it to more people ultimately.

Rob Strechay

>> Yeah, let's dive in a little bit on the isolation thing because, and you talked about breakouts, and I think that's gotten a lot of press recently. People looking at more. People go where the data is, they go and they're looking at this and they're going, people are building inference on Kubernetes. If we can get in there and we can hop out and get to the node out of the container, then we can get across the entire cluster. Is that what a lot of your customer's feedback has been and drove you in this direction?

Lukas Gentele

>> Yeah, absolutely. We've seen customers do virtual clusters at scale, but then separate out the nodes for exactly that reason, right? We've seen customers restrict their tenants really hard. Let's say you want to do image building in a container. You want to have a development container. You want Docker in there, obviously, right? The question is how do you do it without a privileged container? Because all of these things need to run as root typically. Shared environment, you don't want to enable anyone to get these kind of capabilities, right? With vNode, we've wrapped all of these pods on the same node into a virtual node. And my co-founder and CTO Fabian, he has this really cool demo that he runs. He runs a container breakout. He emulates a container breakout, and then he's like, "Oh, we're on the node now. Oh, we're actually not. We're just on the virtual node." Right, you're still stuck in that virtual node. It's very impressive. And I think, yeah, we should probably go back and look at the latest container breakout CVEs over the past year and just try to emulate them with vNode.

Rob Strechay

>> Yeah, that would be a very cool demo to put it mildly. It's great. So Andrew, platform engineering has been all the talk this week. I went to Platform Engineering Day on Tuesday. A lot of it was around how to simplify platform engineering and how to meet in the middle with devs and security? How do you see that and how does your vClusters really help you actually help from an organizational perspective at Aussie Broadband? I'll get that right one of these times.

Andrew Hainsworth

>> Oh, gosh, I don't really think-

Rob Strechay

>> I know I put you on the spot spot here.

Andrew Hainsworth

>> Yeah, that's not a question I've been asked before. But it's absolutely simplified the provisioning element, and I think it's given some of our developer teams who have wanted to get into containerizing our applications and move in that direction. It's increased their confidence in their ability and our ability as the platform to deliver. And having that self-service capability that the vCluster platform has provided has drastically reduced the amount of time that we need to spend holding their hands and that kind of thing. Of course, I don't mean that in a bad way.

Rob Strechay

>> No, I mean but I think that's true because people want to be self-service. I mean, developers don't want to have to put in a ticket and go and open-

Andrew Hainsworth

>> Yeah....

Rob Strechay

>> up things all the time. They want templates. They want things, and they actually want guardrails in a lot of times.

Andrew Hainsworth

>> Yeah.

Rob Strechay

>> They want some amount-

Andrew Hainsworth

>> Template-...

Rob Strechay

>> of choice....

Andrew Hainsworth

>> that's a big thing.

Rob Strechay

>> Yeah. And has that helped you?

Andrew Hainsworth

>> Absolutely. So with Loft Labs and vCluster, we have a concept of an ephemeral case cluster. So part of their pipelines can actually be creating a CAPES cluster from scratch, deploying their application on it. Doing their tests, and destroying that CAPES cluster in an hour and that's just sort of unachievable with a lot of the other methods. Or if it is achievable, it's not as efficient or more clunky, more licensing implications. And so I think from the security perspective, we're still at odds with security in some ways, but yeah, we're-

Rob Strechay

>> But it's getting better....

Andrew Hainsworth

>> it's getting much better.

Rob Strechay

>> And that's a very normal story. I think, again, I call it meet in the middle because again, people want to bring security and bring developers and bring that all together, and platform engineering ends up being in the middle of all of that and trying to bring all of that together. Well, Lukas and Andrew, thank you for coming on board. This has really been a lot of fun. This is something that's near and dear to my heart, and glad to have you here on theCUBE.

Andrew Hainsworth

>> Thank you.

Lukas Gentele

>> Thanks for having us.

Andrew Hainsworth

>> Thanks for having us.

Rob Strechay

>> And thank you for watching this segment here live from KubeCon CloudNativeCon London 2025 on theCUBE, the leader in tech news and analysis. We'll be right back.