Kubernetes Workload Conformance Closes the Portability Gap
Workload conformance is the next frontier for Kubernetes portability. Clusters can already prove they are conformant. The applications running on them often cannot move as easily.
Broadcom Senior Director of Engineering Dilpreet Bindra examines that challenge with Alan Shimel at KubeCon + CloudNativeCon in Salt Lake City.
From Conformant Clusters to Portable Workloads
Kubernetes conformance programs, backed by tools like Sonobuoy, give distributions a shared baseline. AI conformance followed last year. Yet ISVs still struggle to run the same workload across every environment.
Broadcom is co-chairing the new Workload Conformance Working Group in the Kubernetes community. The goal is simple: Kubernetes should be Kubernetes wherever it runs. Shared, verifiable expectations would let vendors validate once instead of certifying platform by platform.
Bindra notes that much of the community’s heavy lifting happens in working groups like this one. Competitors sit at the same table to solve problems no single company can fix alone.
Bringing Virtual Machines Into the Kubernetes Model
The second announcement extends the same thinking to virtual machines. Broadcom built VM Operator to describe VMs through the Kubernetes desired state model. That work now ships as KubeVM Service.
A developer can define a virtual machine in a simple YAML document. Teams can then manage VMs with the same GitOps workflows they use for containers. Rather than keep it proprietary, Broadcom contributed the VM API project to the CNCF Sandbox.
Bindra also points to Broadcom’s longer record upstream. That includes Cluster API work, an etcd debugging tool and the Velero contribution.
AI Raises the Stakes for Platform Teams
Kubernetes has become the de facto platform for AI inferencing and agents. Bindra sees AI compressing projects that once took months into weeks or days.
Enterprises want that speed with strong security guarantees. Broadcom tracks AI usage across its own developers, tying costs to individual features for future budgeting.
For platform teams, workload conformance and a native VM API point toward a more predictable Kubernetes stack. It is one that is ready for whatever AI brings next.
Transcript
Hey everyone. Welcome back here to our live coverage at KubeCon, Salt Lake City. I want to introduce you straight out of Archbishop Molloy High School in Queens, New York, Dilpreet Bindra.
You know some of the football players when they say Dilpreet? Yeah. " Yeah.
And some of them give their high school. So we're going back to high school here for Dilpreet. A couple of Queens boys talking about...
Well, we're going to talk about cloud native, though. Yeah. Anyway, Dilpreet, welcome back.
It's good to have you here. So let's go from Archbishop Molloy in Queens to today. Tell us your story a little bit.
Well, my story, oh, wow. After Molloy, I went to Stony Brook. Good school out there.
Yeah. I lived out there for a while. Then transitioned to working at HP.
Worked on HP UX. The original HP. Is that right?
Yeah. Not the HP now. Right.
I think Dave was still around when I joined. Really? And then transitioned after working on HP UX to working on Solaris at Sun.
And then I've been- That was a company, huh? Yeah. And then after that, I've been at VMware for 20 years now.
Really? This is my 20th year. God bless you.
Yeah. That's unheard of. Yeah.
Unheard of. Yeah. I've been enjoying it ever since I joined.
Yeah. And what do you do now at VMware Broadcom? Well, I drive what we call the VCF workload team.
It drives all of workloads running on top of the platform, including Kubernetes, all of the underlying VM capabilities, as well as now AI. Of course. Yes.
Of course, it's all AI. Yes. Excellent.
Now, our audience knows VMware. They know Broadcom. Yep.
What they may not realize or be aware of is the commitment, the work that Broadcom VMware does with the CNCF. Yep. It's one thing to write a check.
You're a big company- Yes ... you can write a check. Yes.
But it's more than writing a check. Absolutely. Talk to us about that a little bit.
So, over the past decade, we've been a top three contributor to Kubernetes and a top five contributor to CNCF at large. We've contributed to a number of different projects, and we had two interesting announcements this morning. But over the past few years, we've been doubling down on our open source commitments and working on CNCF, like etcd.
We did an etcd debug tool not too long ago. We contributed Valero to CNCF not too long ago. Last year, actually.
Yes. I think that was the last time we spoke. Yeah.
And then we've been doing a lot of work in cluster API, and frankly, we've been doubling down on cluster API more and more and delivering more capabilities there. So all of those things have added up to a lot from a contribution standpoint, and we're not going to stop there. We've got a lot more to do.
The funny part is when we started so many years ago, Kubernetes wasn't this. No. And now, Kubernetes has become not only the de facto platform for applications- Well, for AI as well ...
which it became by last year. Yeah. But now it's the AI stack.
But now it's the de facto platform for AI, both inferencing and for running AI. Absolutely. We did a breakfast summit yesterday with a bunch of panels.
Yeah. One of the panels we did was, actually I moderated it, was The Cloud Stack is the AI Stack. Yes.
And we had SUSE, my friend John Willis, who was here. We actually were supposed to have someone from NVIDIA, but they couldn't do it for reasons we all understand. But that being said, it really is true, and if you think about it, would we have AI the way we have it now had it not been that it was able to just fit on top of this 15-year-old stack that we've built- Yeah ...
with Kubernetes and everything else in there? Likely not. Absolutely.
Because AI is extremely complicated. Yep. The entire stack is complicated, but when you start using the Kubernetes control plane and the mechanisms it provides you, it becomes less complicated.
So that's the cool part. It's funny. Who would've thought Kubernetes makes things less complicated?
Yeah. Anyway, we're going to get into it too, and just say, but before we do, I wanted to bring up another thing. " I know you got the VCF 9 now and all this, and they say it does, but it sounds like you've been doing this for a while.
We've been around for... This is probably, what, my seventh, eighth KubeCon? KubeCon.
And we've been doing a concerted effort around a Kubernetes service for over seven, eight years now. Yeah. So we've been at it for quite some time, and frankly, we have one of the better Kubernetes distributions, I believe.
Agreed. All right. Let's jump to today.
Yeah. You guys made a couple of announcements, and it's not just you're contributing code, you're contributing colon projects. Yes.
Absolutely. Let's talk about today. So today we announced the workload conformance working group.
So Kubernetes as is, through the Kubernetes conformance that we had done with Sonobuoy, as well as the AI conformance that came out last year, Kubernetes clusters have become compliant. But the workloads that run on top of them haven't been easily portable between each of those Kubernetes distributions and each of those individual environments. And what we're trying to do with the workload conformance is to bridge that portability gap so that we work with the ISVs and the workload teams to make sure that they can have the workload run ubiquitously across all these different Kubernetes clusters.
I love it. Before we jump into that, let me close the loop on this one. People who want more information about that one, is there a website already?
Well, we have a Birds of a Feather going on here on November 11th, and I believe we have some material on the CNCF website. It's on CNCF. Well, because it's- It's a CNCF working group.
Absolutely. And if you look under working groups, you'll find this one. It amazes me.
People don't understand. " But you know what? So much of the heavy lifting gets done in these working groups.
Yes. I was talking to my friend Brian at Sonatype, Brian Fox. They got a working group they put together through CNCF now about all the repos.
Yeah. To protect them with- It's impressive what you can do when you bring a set of like-minded people together to solve a problem across various organizations. Across companies.
Yeah. And that's the beauty of the foundation system- Yes ... the not-for-profit.
Yes. Is you could have Broadcom working with whoever, whatever competitors out there. Because at some level, we've got to even out the bad stuff- Yep ...
and let everyone build from there. Yeah, and when we were looking at this workload conformance problem, we were starting to work with all these ISVs, and we were doing a lot of really great partnerships and all this stuff. Sure.
But we always centered our message around the Kubernetes is Kubernetes. Why talk about value? Individual.
Compliance or certification with us, that doesn't make any sense. So how do we make sure Kubernetes is Kubernetes wherever it runs? And look, congratulations to you and Broadcom for recognizing that- Yeah ...
and being part of this kind of effort. Let's talk about the second. So the second one is, again, essentially similar in vein.
We've been working on a API based on the Kubernetes desired state model for our virtual machines for a while. We called it VM Operator inside, and we've delivered that as a product, and now that product is called KubeVM. And what it does, it abstracts all the underlying underpinnings of our virtual machine into a simple YAML document that a developer can use, or any user can use to be able to describe a virtual machine and all the set of pieces that it needs to be able to actually instantiate.
The other thing it does is that now you can start using the GitOps-style workflows that you use for Kubernetes clusters and so on, and containers, you can start using with virtual machines as well. So we did this, and then we were looking at it, and we were scratching our heads, like, why is this just for us? Why aren't we delivering this upstream so that we can have a ubiquitous VM API that is native to Kubernetes?
So we bring the actual native true VM API to Kubernetes. I love it. Yeah.
Where can we get information on this one? This is also run through CNCF. CNCF.
You can find it there. That's a beautiful thing. If you want to learn about KubeVMs, you can come to the Broadcom site or the VMware site.
And if you want to learn more about the VM API project that's being run in CNCF, come to the CNCF website. Love it. Let's talk AI for a minute.
Yeah. It's not really a disruption, but it is. But how is this whole thing playing out in your world?
Yeah. What's the net effect to you and what you guys are seeing and doing? What we're seeing is both ourselves, in our internal use of AI, but also from all the enterprises we work with, we see this dramatic effect of AI increasing the velocity of our enterprises at large.
So what typically would take months is now taking weeks or days. And so we see this dramatic ability to speed up but also automate the capabilities. So, I think we're seeing a lot of attention and push there.
But they want to do so with a lot of security and guarantees that AI is not going to run awry in their system. Right. And rightfully so, because they're all hearing the horror stories, right?
Yeah. And I was asking people today, have you seen ROI? Is it mythical or what?
And you know what? Maybe it's because these are cloud native companies. Yeah.
I don't know if they're really measuring ROI the way maybe you would, but they're definitely seeing, not just at the individual level, but at the organizational level. Yeah. When my developers use AI, we track everything that they're doing from an AI standpoint.
Yep. We're able to understand the cost of what it took to finish a feature. Yeah.
So I can attribute that cost to the feature and know then thereafter I can use that for building out a budgeting exercise to be able to know when I do the next feature, how much of AI cost am I going to expect? Exactly. So.
I love it. Hey, man, anything else we want to share with the audience about what's happening? No, I think it's pretty cool how everything is going on.
The keynotes this morning were awesome. Yes. I was really jiving with the agent substrate keynote, with Tim Hockin and all these guys.
I love Tim. Right. I was telling the guys, look, I remember one year Tim invited us out to Google.
Yeah. Interview him. Yeah.
Google HQ over here. I mean, I've known Tim a long time now. Yeah.
What a great guy. Yeah. And I don't think he had any inkling what this board would turn into all these years later.
Yeah. But God bless him. But I see- He's still wired in ...
his interest is now fully peaked on agents, which makes it really interesting for us to be able to work with him and Google on what that looks like for our customers as well. And again, I asked you, what effect has AI had? So many of my friends, serial entrepreneur types, developers, are re-engaging- Yeah ...
because of AI and what they're able to do and the excitement around it, much more. Yeah. Much more than before.
Yeah. Absolutely. It's an exciting time.
Yeah. I write code myself using AI. You're coding stuff again.
Yeah. And when was the last time you've been in? It was a while ago.
Me and you both. Yeah. I do all kinds of stuff.
Yeah. I make my own AppSec. It's nuts.
Yeah. Anyway, we've got to rock and roll here. Thank you so much.
Thank you. Thank you to Broadcom VMware for everything you guys do- Thank you ... with the community.
I think we might have one more session left for today, day one coverage. We're live. We're at KubeCon.
This is D 极 strong TV.