Open Source Summit NA 2025 | Sean Pomeroy, Kubecost an IBM Company

Clips
News
More from Open Source Summit NA 2025

Sean Pomeroy

Solutions Engineer

Kubecost an IBM Company

Code, community and the future: 13 takeaways from Open Source Summit NA

Beyond uniting developers under shared ideologies, goals and pain points, Open Source Summit North America ignited conversations at the intersection of code, community and change.TheCUBE goes live with insights at Open Source Summit NA

play_circle_outline Maximizing Modernization: The Critical Role of FinOps in Cost Management and Resource Consumption Visibility for Effective Optimization

play_circle_outline OpenCost: An open source project to track Kubernetes container costs effectively.

Info
Transcript

Sean Pomeroy, Kubecost an IBM Company

Sean Pomeroy

Solutions Engineer Kubecost an IBM Company

In this Open Source Summit North America 2025 interview, Sean Pomeroy, solution engineer at Kubecost [an IBM Company], joins theCUBE Research’s Paul Nashawaty to explore the evolving role of FinOps in a world increasingly dominated by Kubernetes, AI and cloud-native workloads. Pomeroy highlights how open source tools like OpenCost are helping engineers and organizations gain early visibility into resource costs, laying the groundwork for smarter modernization strategies.

The conversation dives into the financial side of Kubernetes optimization, touch... Read more

explore Keep Exploring

What factors should individuals consider when starting with cloud cost management and understanding infrastructure costs? add

What is the purpose of OpenCost in relation to understanding container costs in a Kubernetes environment? add

bolt Powered by CUBE AI

Sean Pomeroy, Kubecost an IBM Company

search

Paul Nashawaty

>> Welcome back to the Open Source Summit North America 2025. My name is Paul Nashawaty and I'm covering the app dev space and all things open source. I'm joined today by Sean from Kubecost. Sean, how are you doing?

Sean Pomeroy

>> I'm doing well, Paul. How about yourself?

Paul Nashawaty

>> Great. Why don't you tell the audience a little bit about yourself and Kubecost, what you do?

Sean Pomeroy

>> Sure. So I'm a solution engineer at Kubecost. Been here for going on four years now. Been in the FinOps space for about a decade now, going from cloud cost to Kubernetes, peeling the layers of the onion back. We're here supporting Open Source Summit from a cloud perspective as well as open source. We have an open source project called OpenCost, which is essentially the cost engine behind the Kubecost solution.

Paul Nashawaty

>> Very cool. I mean, financials is a big part of modernization projects. FinOps is a big part of it. I was just at FinOps X recently a couple of weeks back.

Sean Pomeroy

>> Great conference.

Paul Nashawaty

>> Yeah, it was great conference. It was fun. It was good to see all that coming together. It's evolving though, things are happening and changing quickly.

Sean Pomeroy

>> Always. Yeah.

Paul Nashawaty

>> And it's good. I mean, I think that when we look at modernization and we look at the challenges organizations run into, it's always leading into things like costs around that next generation. As a practice lead for app dev, I talk in my practice about past, present, future, and people are looking at moving off their heritage applications to more current state. But a lot of times these organizations, in my opinion, should be looking at FinOps as a precursor to that modernization effort. But in reality, they don't. And they almost have their oh, no factor after the fact and go, "We spent too much money." So at that point, what would you recommend for people that are just getting started in this?

Sean Pomeroy

>> I think it's really important what you said is that most people think about it after the fact. And in our experience, if you lay the groundwork ahead of time, you set yourself up for success much earlier in the game. A lot of the times the issues that we run into are people just don't know what they don't know. So they don't know about the infrastructure. They don't know their costs without some metadata to help them align their cost to different cost centers and business units.

Paul Nashawaty

>> So yeah, I agree. I think that initiatives will help educate the market and we start looking at things like, you mentioned open source, OpenCost. That's an open source approach to help educate the market on these types of things.

Sean Pomeroy

>> Correct.

Paul Nashawaty

>> Tell the audience a little bit about that.

Sean Pomeroy

>> Yeah. So I mean, OpenCost was originally designed as an engine to help you understand from a Kubernetes perspective what the cost of your containers are. If you think about the cloud providers, they're not interested in helping you understand the cost of a subset of a resource. They can tell you how much the virtual machine costs, how much the database costs, but not how much the transaction costs or a piece of that costs. So OpenCost is meant to help demystify that for the engineers to help them get more insight into the actual cost of the CPU to support their application or the GPUs to support their LLMs.

Paul Nashawaty

>> Yeah, that makes a lot of sense. I mean, when you're thinking about the different elements that come with the financial impact of applications, we look at Kubernetes optimization and governance as a big factor too. I think that is, you can't not do some of the things, but you also, you want to do it in a way that's cost-effective so when you're doing it, you don't over-provision. I mean, when I talk to the DevOps community, their business KPI is to get the code out the door, to get product out the door and push. And a lot of times DevOps teams will over-provision because they don't own the budget and they just say, "Look, we need to get our job done." And you can't blame them because they're trying to get their KPIs done. But at the end of the day, the business is sometimes paying 200, 300, 400 times more than they need to be paying. Let's talk about that from an optimization perspective. We look at Kubernetes optimization, how do you optimize at Kubernetes?

Sean Pomeroy

>> Well, it is a really good point because most of the engineers are more focused on performance rather than cost. They want to make sure that their application and the user experience is good, not necessarily that it's cheap or least expensive compared to their other infrastructure. So when we talk about optimization, the biggest thing is understanding what their applications actually need from a resource perspective. And sure, there's a bunch of observability solutions out there, things like Grafana that will give them the insights they need to understand what they're consuming. But then the question would be, well, I see what I'm consuming, but what should I be using? What should my target be? So the algorithms and the processes within Kubecost and OpenCost will help them understand what a more reasonable amount of capacity would be for those workloads based on historical consumption.

Paul Nashawaty

>> Yeah. Well, there's a difference between understanding and seeing and then taking actionable insights. I mean, if you look at and say, "Okay, well hey, you're overspending or you're over-provisioning or you're overdoing," and you go, "Okay. It's still working." But versus that governance of saying, how do you govern those Kubernetes workloads? How do you look at it and go, okay, this workload requires X? And you can be alerted to it, you get to right-size that workload in order for it to be the most effective for your business. Would you agree with that?

Sean Pomeroy

>> Yeah. I mean, definitely right-size the workloads. I mean, we find we follow the FinOps approach, the whole crawl, walk, run when we talk about visibility as well as taking action. So the first step is just the visibility, helping the engineers understand what the cost is or what the optimization routine would be, and then talking about budgets. Am I exceeding my budget? Are the resources more than I should actually be spending? And then ultimately ending up at automation, whether it be just proactively right-sizing those resources as they get deployed, or even to a point where we prohibit or prevent resources being deployed if they're going to exceed a budget based on a forecast.

Paul Nashawaty

>> Yeah, I agree with being proactive on these initiatives. That's what I was saying up front here is when you're looking at modernization, you need to be proactive and look at it. That's ideal. But let's talk about shifts and trends in the market. We're here at Open Source Summit. I'll look back at four or five years ago when the first FinOps X went out in market. I was there and okay, we were talking about sustainability and ESG and all the things around that. And it's like, okay, that was all good. And then this little thing called the AI popped up. And it's like, okay, well sustainability-

Sean Pomeroy

>> Out the door....

Paul Nashawaty

>> out the window and everything that you had to do is now around throwing as much resource as you can. Now that is not going to be the long-term gain. So I think that the FinOps approach is pivoting to these AI ML workloads now and saying, okay, let's take a look at these AI workloads. What can we realistically do to be most effective, not just throw every bit of dollars at it and not build the biggest data center in the world and then not have any power to run it? What do we do? So what are your thoughts on that?

Sean Pomeroy

>> Well, I mean, it's crazy, if you ever look at the retail or the market prices for GPUs, they're anywhere from a couple of hundred bucks to $20,000, $30,000, $40,000. So the question always becomes, are we using the right GPU for our workloads? And then we'll focus on the efficiency of the GPUs, the power consumption, what's actually being consumed from a wattage perspective and from a resource perspective. And then we'll help them make recommendations, whether that be to share the GPU with other workloads or to resize it. Maybe you're requesting multiple cores, but you actually only need one or your workload's not multi-threaded. So helping them-

Paul Nashawaty

>> So fractional GPUs?

Sean Pomeroy

>> Fractional GPUs essentially, yeah.

Paul Nashawaty

>> Yeah. So what would you recommend to the audience? I mean, there's different audience members that are watching here that are going, okay, I'd love to do what you're doing. I've actually been tasked with doing a cost optimization. I don't even know how to get started. What does this mean? I don't want to break the business because if I shut something off. I mean, sure, me personally, I can go in and save any business a lot of money, shut everything off.

Sean Pomeroy

>> Right.

Paul Nashawaty

>> But they'll be out of business.

Sean Pomeroy

>> Right. Yeah. You won't have a very good product. For sure.

Paul Nashawaty

>> Yeah. So how do you get people going in this direction?

Sean Pomeroy

>> Yeah. So I mean, OpenCost will give you the visibility and just to understand what the cost of GPUs are. But when you talk about efficiency and optimization, a tool like Kubecost, an enterprise or a paid offering, we do have a free tier, will give you those insights and help you understand on a workload perspective how they're consuming and what changes can be made. Whether that be changing the GPU type or the machine type to make better use of the different GPU capabilities that are out there.

Paul Nashawaty

>> Okay, that's a good approach in the start. When I think of these different approaches, how do I know which approach is the right approach? Because I think when I looked at FinOps X, there was 88 vendors there?

Sean Pomeroy

>> Something like that.

Paul Nashawaty

>> So I thought it was huge. But I mean, I know that something is looking at rates, something is looking at utilization, something is looking at... And there's different things. If it's a little bit overwhelming, what are some of the baselines that people can start with?

Sean Pomeroy

>> Yeah, it's a good question because I think we're early enough, and like you said, we're at a tipping point with as much as the buzz is around AI that we're still developing the processes that make sense. And I think each business unit, each team has a different approach to it. So my recommendation would be to see what's out in the market, see what makes sense based on what your goals are as a company based on the workloads that you're deploying.

Paul Nashawaty

>> Yeah, that's a fair response. I mean, I appreciate that. I think that you meet the client where they are on their journey and understand how they're achieving it. What are some of the, I would call it the tips and tricks that you can share with the audience and say, hey, this what I could offer up?

Sean Pomeroy

>> Yeah. I mean, tips and tricks are tough too because a lot of the problems right now are availability of the resources. There's a lot of regions in AWS where you don't have availability for that. So a lot of people we see are actually using the resources that are just available to them. May not be the right ones for the workloads that they need, but it's all they can get right now.

Paul Nashawaty

>> Sure.

Sean Pomeroy

>> Tips and tricks. One of the biggest things we run into in Kubernetes, maybe not necessarily specific to GPUs, is not setting requests and limits on your workloads.

Paul Nashawaty

>> Okay.

Sean Pomeroy

>> So when you deploy things to Kubernetes, if the control plane doesn't know what you need from a resource perspective, you can run into performance issues that can ultimately lead to a run on cost.

Paul Nashawaty

>> Okay. That makes sense. Would that be, would you say, something different for pitfalls?

Sean Pomeroy

>> Yeah. Maybe not giving much capacity to the workloads. But I think the important part here is governance, is basically setting the guardrails to ensure that people are actually adhering to what they should be when they're deploying their workloads.

Paul Nashawaty

>> Very true. So if we're here a year from now, when we're sitting here, we're having this conversation at the next Open Source Summit, where do you see the market going? What would you want to see happen in FinOps? And what would be the change?

Sean Pomeroy

>> I'd certainly like to take a step back to go back to sustainability because I think it's important that we continue on that path, especially with AI, because we've seen the power consumption go through the roof. They're spinning up nuclear power plants in order to support these GPU workloads and we're removing our focus on sustainability. So I'm hoping we get back to that.

Paul Nashawaty

>> Yeah, I think you're right. I think we're going to have to normalize that. Once we normalize and harmonize what we're actually using these resources for, that's going to help dictate the financial spend. But this has been great. I mean, I really do appreciate your time today. There's a lot going on. Final word to the audience?

Sean Pomeroy

>> Oh, man. I mean, specific to open source, there's a lot of really cool solutions out here today. I've taken a walk around. There's a lot of stuff around AI and LLMs and being more efficient with that. And then I see OpenSearch right here, they've got a really cool solution. So I'd say see what's in the market, take a step back, understand what solutions are there, and find what fits you best.

Paul Nashawaty

>> Very cool. Thank you for your time today. I really appreciate your insights. It really is important to share this with the larger audience. And I want to thank you for watching today. My name is Paul Nashawaty and I'm coming to you live from the show floor at Open Source Summit North America 2025. And thank you for watching theCUBE, the leading source in tech news.