Reducing the Carbon Footprint of K8 Clusters with PerfectScale’s Amir Banet

This is Textron tv. Hey guys, thanks for the throw. We're here with Amir Bennet, who is CEO for Perfect Scale, and we're talking about how to reduce the carbon footprint of Kubernetes clusters, of which there are starting to be a lot of them in various environments.

Amir, welcome to show. Thank you very much, and thank you for having me. What exactly is the challenge here?

We've been struggling with carbon footprint for all of it, but is there something unique about Kubernetes specifically that is more challenging than other platforms? Yeah, I think that the nature of the beast, uh, talking about Kubernetes as an ephemeral system that constantly changing makes it much more complex to really understand what it's, uh, either cost or even the carbon footprint. Uh, so all of, uh, again, the relevant metrics are much harder to collect and to aggregate because we're talking about a system that's constantly changing and the amount of nodes, uh, and the amount of pods is directly affected by the actual, uh, demand that you have from your MDA customers or end users.

And this demand is also fluctuating. So it is much harder, again, to predict or have any analysis about what is the CO2 consumption of your environment because of the nature of this beast. How much pressure is there now to kind of measure all this stuff?

And will we need to do this alongside everything else we're measuring, or is it gonna be more of a separate motion? I mean, it seems to me we already have a lot of tools for monitoring various things. So can I just Mm-Hmm.

Put this in my workflow? It's a great question. I, I think the answer is depending on the size of the organization and the location of the organization, and that is because we know, uh, that in Europe there is a much bigger kind of, uh, pressure or demand to have these, uh, KPIs or this information available because they're actually being measured, meaning the company itself is being measured on the impact that they have on the environment.

Either the negative impact or the positive impact. There is also, uh, kind of, uh, uh, incentive programs for companies to reduce their negative impact on the, on the environment and even to to pay fines if this is not, uh, uh, being done appropriate. And it all starts with the oversight, whether you have this kind of a visibility to begin with.

And from there, we can talk about how can you optimize, how can you, uh, reduce the amount of, uh, again, uh, CO2 consumption, for example, uh, that your company, uh, consume. Uh, so in Europe, it is very hot trend. I believe it is only a matter of time until, uh, it'll also catch up in, uh, north America and the rest of the world.

Now, whether or not it can be as part of the monitoring solutions, well, it depends on whether the traditional monitoring solution will do any activities in that regards. Uh, so far I haven't seen any, uh, real progress there. Uh, and this is why we are so excited about what we are doing in this space, because again, it's not our kind of, uh, main bread and butter, and we can talk a little bit about what perfect scale is all about, but this is a very, uh, kind of important steps that we would like to add to our value propositions that also makes us unique in this space.

Well, let's go there. What exactly is the whole motion around governance and optimization of Kubernetes clusters these days? I mean, I think a lot of folks are kind of now struggling with that after managing onesie twosie.

So where are we on this journey? Yep, great question. So we are exactly in this, uh, governance and optimization space in Kubernetes.

And what we have is the AI driven, uh, solution that autonomously and continuously help you to find the right, uh, balance point between two contradicting forces. One is the, uh, the performance, durability, resiliency, and the other one is the cost. So no matter what aspect you are talking about in Kubernetes, we are able, again, to, to give you the specific recommendation of what should be the values to be used in numerous amount of parameters, uh, that can start with the, the pods or the containers, and what should be the right size going forward to the nodes, which nodes, uh, should you select?

Uh, and moving on to things like HPA, how to set the right trigger point to have more replicas or less replicas. So all of these points are today are being guesstimated by either the dev, the DevOps site reliability engineer, and we want, again, to give them a data-driven, uh, solution that helps them to stop guesstimating and take, again, a very scientific approach to this. Uh, and this scientific approach can be either, again, continuous optimizing on your behalf, your clusters, or just to provide you the recommendation.

So this is the optimization side of the house. The other side is the governance. Here we're talking about, again, giving you the full visibility and control to whatever level of granularity that you care about in the Kubernetes, which is considered to be a black box to some organizations that are not able to really understand what's going on either in the cost or in the waste or in the amount of resiliency issues that they have.

So we give them, again, this, uh, single, uh, pane of glass to really understand what is going on, and more importantly, how to fix things that are currently not, uh, appropriate in their clusters. One of the perceptions people have, especially when it comes to either cost or to carbon footprint for that matter, is that mm-hmm, theoretically, Kubernetes clusters enable you to dynamically scale up and scale down, right? It doesn't seem like we're scaling down that often.

We're over provisioning. So what's going on there? So, as you correctly mentioned, Kubernetes is a very powerful, uh, open source.

And one of the reason for that is because it has so many built-in functionality as part of it. Uh, and those built-in functionality also includes the vertical product scaler that is helping you to again, find the right size of the pod and also the horizontal port autoscaler that allows you to have more replicas of the same pod in order to, again, to address, uh, the, the moral load that is coming. However, these, uh, components have a lot of advantages, but also a little bit disadvantages or shortcomings.

Uh, and this is why we created our own, uh, version of VPA, uh, that, uh, has, uh, all the advantages that VPA has, but also able to address the disadvantages that it has, and it can run very smoothly together with your HPA, no matter if you are using keda or whatever. Um, so again, we kind of, uh, provide either alternatives to the, uh, engines that Kubernetes come with, or we are providing kind of a, a brain on top of HPA, for example. So it'll function much more, uh, optic.

Uh, so again, yes, the machines are there, but it's still those machines that are coming with Kubernetes needs a lot of, uh, handling and a lot of, uh, guesstimating. And we want, again, all of this operation to be as efficient and as, uh, a prone, uh, less as possible. So this is where, where our platform comes, uh, in handy.

You mentioned AI earlier. Where does that play in this equation? I mean, I'm assuming there are machine learning algorithms involved, but there's also a lot of talk about generative AI these days.

So where are we on that journey? Yeah, so we are tailoring our recommendation to what is exactly happening in your clusters. So we do have algorithms, but these algorithms are learning exactly how each cluster behaves because we know that, uh, one cluster can be completely different than another.

Uh, and based on this learning, we are able, uh, to get, to give you the exact recommendation that takes into account also the most, uh, surreal, uh, kind of edge cases, uh, or niche, uh, niche cases. And, uh, we are again able to do, uh, the best recommendation in the market because our AI build, uh, algorithm that is focusing not only on statistical behavior, but also real understanding about how, what are the different, uh, again, uh, edge cases and how each H case should be addressed, depending again, on what's going on in the customer environment. Who makes the decision around the whole carbon footprint?

Is that coming out of an IT leadership, or is there other people involved in driving this conversation? So, as I mentioned, in Europe, at least, this is a mandate, uh, that is, uh, either follows on the chief data or the chief information, uh, officer, uh, because again, the, the need to provide this visibility and to get even the, the badges or the certification that are needed, uh, is, uh, uh, a top down kind of an approach. So it comes from the CEO or the, the other c executive, and then they're requesting, uh, to the head of DevOps or the head of infrastructure, uh, basically the numbers or the, the data that they need in order to qualify for these, uh, certificates, et cetera.

So like very like, uh, you can see in the monitoring space, the, the need for cost efficiency, uh, is coming from top down, but in the end, the people that needs to provide the information and also to help with the optimization, uh, the practitioner, uh, in, in our case, uh, because again, we in, in perfect scale, we provide not only the visibility to what is your current carbon footprint for which one of your clusters, but we are also a able to help you to reduce it. And, uh, work for the reduction is, uh, done by the same people that are doing this work on the cost, uh, perspective. So as you are, again, focusing on reducing your cost in the same kind of a breath, you can also be focusing on reducing your carbon footprint, uh, uh, emissions.

Do you think that this is all gonna be driven by incentives or will there be penalties for exceeding carbon footprint and what do they look like? Yeah, that's a great question. I think it's, uh, going to be a combination of, uh, the stick in the carrot.

So, uh, and I think that there will also not think that, I know that there is also a market of, uh, selling those, uh, incentives or certificates or credits. Uh, again, it depends on each country, uh, between different, uh, organizations. Uh, so for example, if you are very, uh, positive in this regards and you have credits that you gain, you can even sell it to other companies, uh, that are in, in shortage.

Uh, so yes, there will be more a kind of, uh, control coming from the governments on this space and organization will need to align to this new, uh, regime. Uh, again, by having both awareness to what is their current carbon footprint and also kind of a plan if they are, uh, currently not, uh, very, uh, let's call it, uh, friendly to the environment, how to improve this, uh, moving forward, What's your best advice to folks about how to get started with all this? 'cause some folks will just do it out of they wanna, you know, participate in a green program, so they don't necessarily need an incentive, but should I create like an office of the Green IT folks, or how do I kind of pursue this whole thing?

Yeah, that's great. So again, we are not, uh, claiming to, to be, uh, a carbon footprint monitoring across the board. Perfect scale is only providing the, the focus on the Kubernetes side of the house.

Uh, I can tell you that, uh, what we have done, uh, couldn't have been possible, uh, if the community, uh, was not also in our side here. So, uh, we are, uh, our solution is built on other open source, uh, uh, solutions or companies that are providing value here. So I do recommend you to look at, uh, uh, few, uh, website.

One of them is ds, uh, engineering, and the other one is the carbon footprint, uh, dot com. So in there you can find a lot of information and also calculators to see exactly how much you are currently, uh, consuming. And this is broader than just Kubernetes.

So for example, LI C two, for example, serverless, all of this is being part of, uh, uh, the open source, uh, initiatives. So again, it's something that, uh, definitely requires some research from your end because there is no, as far as I know, there is no single solution that can provide you all the different carbon footprint that you have in your organization across the board. Uh, but I think that in the future, yes, you will find dedicated solution that this is what the kind of sole purpose is.

All right, folks, well, you heard it here. Regardless of whether or not you believe in climate change, it doesn't really matter because, well, it's gonna become a requirement. So you gotta figure out how you're gonna deal with this carbon footprint issue no matter what.

So it's time to get started sooner than later. Hey Amir, thanks for being on the show. Thank you very much, Michael.

Sure. Having you here. All right.

And back to you guys in the.