Become a Creator today!Start creating today - Share your story with the world!
Start for free
00:00:00
00:00:01
Green IT on a Cloud: Talking sustainability, seriously
6 Plays1 day ago
Welcome to Cycloid's Green IT on a Cloud, the podcast show from Cycloid where we turn sustainability from a buzzword into a to-do list.
If you’ve ever stared at a cloud bill that reads like a plot twist, you’re in the right place.
Promise: fewer acronyms than a KYAML… and when they do sneak in, we’ll translate. For example, PUE = power use effectiveness; WUE = water use effectiveness - but you’ll have to wait to hear the question!
Your host Ben Hewison is joined by two brilliant guests:
Benoit Petit from Hubblo, open-source powerhouse behind the Scaphandre energy-measurement agent and long-time voice for evidence-based Green IT. He helps teams measure what matters and ditch what doesn’t.
Julien Syx, CTO & Product Lead at Cycloid, he builds platform-engineering guardrails so the green and golden path is the easy path for developers and ops. Think: self-service, good defaults, and less waste sneaking into production.
What’s on the menu for greenifying your day on this episode?
- Data centers & growth: when expansion helps, or hurts, energy and water.
- Reading the fine print: what to ask about PUE/WUE, cooling, and disclosure.
- AI reality check: cutting compute, cost, and water without killing innovation.
- GreenOps × FinOps: aligning budgets with impact - no, they’re not the same.
- Your 90-day plan: three moves you can ship this quarter.
Enjoy the episode, and find out more about sustainability, Green IT, FinOps, GreenOps, Internal Developer Platforms and Portals on www.cycloid.io
Recommended
Transcript
Introduction to 'Green IT on a Cloud'
00:00:11
Speaker
Hello and welcome to Green IT on a Cloud, Cycloids podcast, the show where we turn sustainability from a buzzword into a to-do list. I'm Ben Hewison, your host.
00:00:23
Speaker
If you've ever stared at a cloud build that reads like a plot twist in a thriller novel, you're in the right place. Here's a promise for you. We're going to offer fewer acronyms than KEYAML, and when they do sneak in, we will translate.
00:00:36
Speaker
So for example, PUE is power use effectiveness. WUE is water use effectiveness, but you're going to have to wait to hear that question.
Meet the Experts: Benoit Petit and Julian Six
00:00:45
Speaker
Today, I'm joined by two brilliant guests. We have Benoit Petit from Hublot, an open source powerhouse behind the Scapandre Energy Measurement Agent and a long time voice for evidence based green IT.
00:00:57
Speaker
He helps teams measure what matters and ditch what doesn't. And I'd also like to introduce Julian Six, the CTO and product leader at Cycloid. He builds platform engineering guardrails, so the green and golden path is the easy path for developers to follow.
00:01:12
Speaker
Think self-service, better defaults, and less waste sneaking into production.
Key Topics: Data Centers, AI, and Sustainability Plans
00:01:17
Speaker
So what's on the menu for greenifying our day? Well, we're going to look a little bit into data centers and growth.
00:01:25
Speaker
We're going to look at reading the fine print when it talks to things like PUE and WUE, cooling and disclosure. We'll take a reality check on AI. Can we cut compute cost and water without killing innovation?
00:01:38
Speaker
And of course, we'll look at Green Ops versus FinOps, right? Or with FinOps, right? So how can we align budgets with impact? Because they're not the same thing. And then of course, we'll also look at a 90 day plan.
00:01:51
Speaker
What three things could you do this quarter? Or what thing could you do these next weeks to help you improve your sustainability? And yes, we will keep the jargon in staging. So if anyone says synergy, They owe the show and the world a tree.
00:02:05
Speaker
Ready? Let's dive
Benoit Petit's Work at Hublot and BoaVista
00:02:07
Speaker
in. So first of all, I'd like to ask you, Benoit, just introduce yourself and let our audience know a little bit about yourself. Over to you. Hi, thank you for having me. So I'm Bonapetit, I work for Hublot, who is a study the company running studies for governmental agencies, clients from the private sector as well. And every time it's about ICT and environmental questions. It could be on different aspects.
00:02:41
Speaker
We do, as you said, a bit of software too, but we are not a software editor. It's more like we build software when it makes sense to... highlight impact outputs, let's say, impact biggest shares on a given system, or to highlight big macro tendencies to make people and just understand better what ICT environmental impact means. um
00:03:14
Speaker
And we help organizations to transform themselves as well. It's not just technical.
00:03:21
Speaker
It's more like general purpose advice depending on what the person in front of us is doing in the company. And we also, i just to finish, i I also work as a volunteer in a nonprofit called BoaVista.
00:03:40
Speaker
ah which is producing digital commons to help with this topic. So better understanding, better measurements of different kinds of ICGN impacts.
00:03:54
Speaker
So yeah, that's that's ah that's all for me. Brilliant Benoit. Well, fantastic. Thank you. Interesting life and a busy life and lots of activity in this green and sustainability space. That's fantastic to hear.
00:04:06
Speaker
um
Julian Six on FinOps and GreenOps at Cycloid
00:04:07
Speaker
Thank you. OK, so um now, Julien, if you could just introduce yourself, let us know a little bit about your background, and and yeah, again, anything that you'd like to tell the world.
00:04:16
Speaker
Sure. So hello, everyone. I'm Julien. So I have an ops background. ah But when I'm not doing ops, I'm doing some development. And when I'm doing too much development, then I get back on ops because, yeah, you know are always balancing between both. yeah, I'm so yeah um um I joined Cycloid more than 10 years ago, founded it with Benjamin Rial. So at Cycloid, we are building a ah platform engineering that is unified also with an internal developer platform, so you have the best of both worlds mixing together i know in one product.
00:04:58
Speaker
So you may wonder how it is related to sustainability and mean why I'm currently in this call. And that's a good question. We integrated a FinOps and GreenOps module in our product because we think that helping centralizing all things you deploy on cloud and being able to track it because we found that there's so much waste on money, on cost, but also on environmental cost. on cloud providers because people are deploying things and they totally forget, but they deployed something. So we think that centralizing everything into the new world platform and having a way to track all the assets from all the different cloud providers that you connect into one platform will help to to to produce some reporting and being able to track, as I said, the financial costs, but also
00:05:54
Speaker
the environmental cost is trying to estimate. I think we'll talk during these ah this podcast about how estimating things, because that's a really complicated topic, but at least we try to give an estimation.
00:06:07
Speaker
And um so, yeah, that's why I'm currently in this call. And also on the on the personal level, I'm a climate risk facilitator, trying to to spread the some awareness about the sustainability and so so on.
00:06:21
Speaker
So that's it. Fantastic. Thank you. Well, as we move forward, we will try to balance the load of questions that we ask and make sure that we focus on finance and the ah the sustainability element, because they do go hand in hand. Right.
00:06:34
Speaker
And um as we know, it's important to improve our bottom line as we improve the Earth's bottom line. So without further ado, let's dive into some questions. We're going to use a Q&A format. Right. So I will first ask Benoit question and then open the table for Julien to have ah you know to intercede with his thoughts as well.
00:06:51
Speaker
But of course, it's an open forum. It's just a chat and let's go for it.
Is Optimization Enough? A Three-Tier Framework
00:06:55
Speaker
So um as Benoit is our special guest, and I will, of course, give him the honor of answering our first question.
00:07:02
Speaker
um So Benoit, what is the biggest myth about green IT t and what data or what kind of process or or what reporting disproves that? Right.
00:07:16
Speaker
I'd say the biggest one would be that it's enough to optimize in Green IT, in ICT in general. That's the the the first focus most people have.
00:07:29
Speaker
And while it can be useful, and sometimes it's very useful, I'm not saying the opposite opposite, it's really like the third kind of impactful action you might have.
00:07:43
Speaker
um So the idea very summarized is that we could split, um let's say, environmental action in ICT in three categories.
00:07:54
Speaker
It's a framework from me, so it's not like the guidebook for it. It's just a way to think about it. So you would have oh you would have first robustness actions or business model redirection actions. So thinking about really what you are producing and if it's compatible with planetary limits, and if not, how you could adapt that, which means way more than just technical questions. It means organizational questions, business questions, and so on. So it's everything in a row.
00:08:33
Speaker
um Second one would be what in France we call sobriety, but that this term doesn't fit much with Outside of France, it doesn't work that much. i think we should find another word. But anyway, sobriety would mean thinking about your impacts as absolute impacts. So thinking about the impacts of the whole company and the whole activity, not just the subsystem.
00:08:57
Speaker
Even if after you have identified very impactful parts of your activity, you might focus on it and then focus on the system and then optimize. That makes sense. And then you have optimization. And most of the time what happens is that we go through optimization, then realize it's not enough, go through flu value three, then realize it's maybe not enough yet and go through redirection of business model thinking. But I think if we started on the other way, it would work much better.
00:09:29
Speaker
So yeah, but just to prove that, and then I will let Julien answer as well, to prove that optimization is not enough, that's, as I said, the 90% ninety in the ninety ninety percent cases, we go through optimization first.
00:09:45
Speaker
The industry, the ICT industry says we are optimizing, we are doing better efficiency. Everything is about that and the results are here. All the big techs have increasing greenhouse gas emissions, yearly greenhouse gas emissions by double digit figures.
00:10:04
Speaker
um ICT is still one of the sectors with the fastest pace of greenhouse gas emission growth. I think it's like six or seven percent per year, or something like that.
00:10:18
Speaker
um Data centers will consume twice as much as today in 2030, most probably. so everything shows that it just doesn't work. We have to rethink the whole framework, not just optimizing system. That that would be my take on on this.
00:10:37
Speaker
okay Yeah, it sounds like we have an optimization and a process problem, right? And then also, you know, in effect, which leads us to positioning issues, right? um Julian, I heard the word that I know that we use a lot in cycloids, sobriety, right? How how do you feel about that?
00:10:51
Speaker
ah Yeah, but I can't agree more about ah about the issue with optimization. And I always, because I use it also this argument during my climate-first animation, I always take some comparison with cars.
00:11:10
Speaker
because ah cars, I think everyone has a car and everyone can understand this comparison. We always try to improve and optimize cars to reduce the consumption. And finally, you had cars 20, 30, 40 years ago that were consuming, i don't know, like five liters per hundred kilometers. And now we still have exactly the same, but we just have bigger cars.
00:11:34
Speaker
And finally, we didn't optimize, ah we optimized, but we didn't reduce the consumption. We just ah growth the whole car and so on. And finally,
00:11:45
Speaker
that's a kind of a rebound effect of optimization we are always increasing increasing increasing and even if you try optimizing the the emission about i don't know gpu and so on it will just be to add more and more gpu and finally staying exactly at the same level so i totally agree that optimization is the wrong way ah to start with and would be better to first thinking about decreasing and then optimizing at the end Yeah, brilliant. I mean, I think it it encapsulates a lot business thought, right, that sometimes that easier kind of more KPI influenced path seems something that the better path to take.
00:12:24
Speaker
And then when we end up with ourselves, be and we see what we've gone wrong. um Brilliant. Thank you very much, gentlemen. OK, so on to the next question. um And Julian,
Challenges of Data Center Expansion
00:12:33
Speaker
this one is for you. So I'm going to talk about something I think fairly close to your heart in some way.
00:12:37
Speaker
It's data center growth, right? But when does expansion not hurt sustainability. And is that phrase even possible in reality? Or does physics tell us no?
00:12:49
Speaker
Yeah, I don't think it's really a visible. I think it's ah it sounds a bit paradox paradoxical. I'm not sure about ah expansion considering the state of the planet could go with ah sustainability. We can probably mitigate by increasing more and more efficiency of data centers using renewable energy and so on. But I don't think We can talk about sustainability ah because as I said previously, i say it's like Carlson, we're just adding more and more and more and expansion is trying to to expand something infinite into finite world. so so so So, yeah, I don't think that a center will be able to grow because it's not only about energy because we are we are really
00:13:38
Speaker
but we can always focus on energy, but on the data center, there's also the water consumption, but also ah your your you're using the the land ah because you will need some concrete in order to to create the building and so on. You will need a lot of materials like iron and all those things to to to create data center.
00:14:05
Speaker
So it can't grow forever and ever considering all the planet limits. And I'm not only talking about the climate. ah I don't think we can talk about sustainability if we talk about expansion.
00:14:21
Speaker
addition I think I have to agree. Benoit, what do you think on this level? Is there any possibility for expansion with sustainability? Is there any way to optimize or is it just the a pipe dream?
00:14:33
Speaker
ah yeah It links to the the first question, like, if are expanding while optimizing, yeah, you might reduce per data center impacts that the idea is how many data centers and how big they are when you build them.
00:14:47
Speaker
And the trend, it's very, very interesting this is because ai Wave is just enhancing all the trends that were already there, but by sometimes 10 times.
00:15:00
Speaker
And so it's like even more obvious than it was before, even if the industry is very ah hiding some figures. and And yeah, that's why microsoft got plus 30% on greenhouse gas emissions in one year.
00:15:16
Speaker
It's mainly because of concrete of the new data centers. um and while So what would work, to be very, very honest about the situation we are in, what would work is densification, keeping the same building in hardware capacity.
00:15:36
Speaker
That would work. if we focus on the same capacity, keep it, then optimize to make as many use cases as possible fit in that capacity.
00:15:50
Speaker
that would be effective um That would be effective action, but it would mean planification, it would mean maybe be regulation, maybe quotas may be limiting the resources we allow ourselves to use. And that's really the opposite of how it works in ICT today and how even the countries are acting regarding ICT, because these days are more about deregulation. It's more about, in France, we have those fast-track projects, which means we will bypass all the regulation steps to be able to build new data centers. In the US, it's even more.
00:16:35
Speaker
so We saw it with the second version of the Colossus data centers, which with gas power plants all over the place, with the electricity production put just across the border of the nearby states to be able to bypass the environmental restrictions. So everything is about deregulation today. So basically, I don't think it would work, but that's the way we are going. So we'll see if it works at some point. Maybe that will prove me wrong.
00:17:09
Speaker
Yeah, well, I think the environment and humanity can hope that we are all proved wrong on that front. and let's see Let's see where we go. Of course, we can't predict the future. But maybe we could predict what happened in the next
Improving Sustainability through Relationships and Training
00:17:21
Speaker
90 days. So Benoit, this next question is for you.
00:17:24
Speaker
um I'm going to call it a 90-day plan, but it's just a 90-day thought process, right? if If you could only do you know three things at maximum or even one thing in the next 90 days, what is it or what are they?
00:17:36
Speaker
and And what you know green washing action would you refuse to take?
00:17:44
Speaker
So in the next 90 days, um in such a short period, I would focus on the relationship with the manufacturers and vendors are relying on and operators could be also the test and operator.
00:18:00
Speaker
um and set rules that are mandatory so that I, as a company, would buy their services or products. And it would focus those rules would focus on hardware, lifetime hardware, software, compatibility with all the hardware, maintenance of software so that my hardware doesn't go to waste before, let's say, 10 years. That would be good.
00:18:30
Speaker
So it's far from the reality today. In the workplace, the average lifetime of devices is something like four years. For servers, could be six years, but that's not much more than that.
00:18:46
Speaker
um So that that's an action that Many people would say it's hard to take. I would say it's one of the easiest because at some point, you are a company, you have a lot of money, you can buy a lot of equipment to a manufacturer, and you can force them to set different standards.
00:19:06
Speaker
And the second point in 90 days would be training and um explaining ah to every stakeholder, including people from the board,
00:19:19
Speaker
in what situation we are, and then to kind the the right actions might have a better chance to come out if you explain exactly the situation where in not just the usual green rushing stuff would be, it links to your second question, green rushing to avoid would be humanity will adapt.
00:19:42
Speaker
That's OK. It would be our company is doing its best because we can count on one hand the companies in the world that are actually doing very ambitious.
00:19:54
Speaker
impact reduction plans. not Not more on one hand, really. ah and And especially in ICT, what I would avoid is what we do, is what the industry does every time, focusing on the relative impact of one unit.
00:20:13
Speaker
Focusing on, okay, I reduced, or here is, the i won't mention anyone, here is the impact of one inference on my JNAI system.
00:20:25
Speaker
ah This is big issue. the The topic is not what is the impact of one inference. The topic is what's the impact of the whole system and what does it mean locally, globally, on different impact criteria. Those questions are never um addressed. And if you don't address them, you're just not transparent. So that's a great question.
00:20:47
Speaker
that That makes a lot of sense. say Uncomfortable truths make people uncomfortable. And if they can avoid talking about those uncomfortable truths, they always will do. Right. And ah we kind of accept that publicly, I think a lot of the time from individuals and from companies as well. And also from, of course, from entire industries and countries.
00:21:04
Speaker
Right. So that's a very key point. be well I love that one. um Julien, what do you have any thoughts on this one as well? No, 90 days. what What could we do in a kind of technological sense or in a coding sense or in a programming sense to help ourselves in 90 days?
00:21:17
Speaker
Yeah, totally. I have a cycloid basis. I will preach sure little a little bit what we're doing at cycloid. But yeah, on an IT day, I will especially focus on the on the shadow of IT and track all the things that are running.
00:21:34
Speaker
and shouldn't and are trying to automate start and stop for non-production environments and so on in in the company. It's kind of related with the the first question when we're talking about sobriety and I will start by that. And I think then on on the next day, i totally agree with Benoit. I will try to to to to enforce with some customer power, let's say. to infer some policies ah about put out the hardware.
00:22:05
Speaker
um And I will try to to enforce policies across ah all the company ah to ensure that we we stay a bit more aware about um but having I think some policies to avoid shadow IT, to consume with ah being more sober and also take care about the hardware we are buying and how could we use it until the end.
00:22:33
Speaker
i mean, if it's still working, if it's still ah adapted to to a work, then let's try to to to use it and not just replacing things because there's something that is fancier and better. So and yeah so but but but that's what I will do. i will try ready to... The greenwashing I would refuse is right to to buy a also a cheap ah cheap carbon offset to to claim carbon neutrality, while in your ring that all the cloud waste already have, I think a priority must be to reduce the consumption first. That's brilliant. yeah you know and Embracing putting light on the shadow IT by avoiding buying shiny things is a very nice, neat message.
00:23:22
Speaker
I think that's a great one. and And I think it really does matter as well. I've read a a Reddit post post recently about instances that were spun up and forgotten about for six, seven months. And you know they were losing thousands of euros every month, but also all that carbon, all of that electricity, all of that water, everything just wasted for nothing.
00:23:38
Speaker
So I think it's it's really important that we maintain an understanding and look into those dark spaces that we don't want to talk about and that we are not willing to share publicly. um Next question for you, Julian.
00:23:51
Speaker
It's an interesting one. I think it follows
Energy Consumption: Programming Languages and Data Transfer
00:23:53
Speaker
on from that a little bit. right It's this idea of um from code, what we code, into what energy and what electricity it uses. So, I mean, are there software choices and coding choices that we can make that have real impact, you know such as the language which we code in,
00:24:08
Speaker
the amount we cache, you know data size, you know or this idea of batch versus real time. you know Are there software and programming choices we can make that would improve our consumption or reduce it rather than improve it, I guess?
00:24:24
Speaker
Yeah, I think there are few few few things you can you you can change on on the software choices. The first one is the programming language. There are a few studies, scientific studies that exist about ah the difference between computer languages and scripting languages.
00:24:46
Speaker
We can observe, like for example, that on C, Rust, C++, plus plus you have a 50 time factor ah compared to to scripting languages like Python or Perl.
00:24:59
Speaker
ah and So yeah, computer languages are usually more energy efficient so for the same computational task. so So I think you you should consider ah sos a the product you're doing, or at least the program you're doing, if you if it could be better with Compil language.
00:25:24
Speaker
I think there are also by all the data transfer also. ah Which technology do we want to use? we There are also some studies that prove that using a cellular network,
00:25:38
Speaker
It's 10 times more energy consuming than a V3 connection. so So yeah, we talk about ah we talk a lot about the 5G revolution, about yeah IoT, etc. I think i think we could have a some focus also on the data transfer.
00:25:57
Speaker
and now And also the famous batch versus real time. ah There are also some kind of improvement where you can shift the workload time when this the grid is greener.
00:26:11
Speaker
ah So doing some what we call carbon aware computing. I think i think you can it can use sometimes focus on those kind of improvements. If you have some some ingestion, for example, of patch and so on, you can do it at night. ah ah but You can also do it on the day, but thanks to the cloud, you can find some some cloud ah some some data center that are at the opposite of the world.
00:26:45
Speaker
And they're doing the night, they're doing, a ah they're using solar energy or well, anything. you You can reduce some consumption, I think. Brilliant, great. Thank you, Julian. Benoit, what do you think? what ah What software choices or coding choices or, you know, language choices can we make that would ah that would impact um positively this use of electricity and water, etc.?
00:27:16
Speaker
um yeah on On the technical aspect of it, it's potentially a complex complex topic because yeah for sure if you compare in a very simple kind of task compiled language to interpreted language, you might have better results. yeah that that yeah Actually, agree there are studies about that. withs there There is a project, ah actually an open source project that's ongoing in BoaVista on trying to get get some insights on that. But what they tried to integrate was not just the choice of the language, it was also the choice of the kind of database technology, whether there were cages or not, ah whether the the code was pretty optimized or written, let's say, in a rush, not very optimized.
00:28:08
Speaker
They try to integrate several variables like that, which doesn't give a complete understanding of the topic because, of course, it's a complex topic, but it gives some insights on would we get better results in terms of hardware weight or environmental impact.
00:28:30
Speaker
on this stack or another. And it's very interesting because it shows that when you add up variables to the equation, it's not as binary as we can imagine.
00:28:42
Speaker
Let's say you could have a stack with Rust that would be worse than a stack in PHP. depending on what are the other variables and how. Yeah, it really depends, but generally speaking, I think if you had to develop the Linux kernel in JavaScript, I'm yeah pretty sure you'll have some weird results.
00:29:01
Speaker
Of course, the question is what what ah what is the use case you serve? And it was, to to be fair, this this Let's say this open benchmark is more about ah web apps.
00:29:12
Speaker
So it's one kind of use case. If we go on another use case, we will probably have different results. And for sure, if you talk about code that doesn't have so much dependencies and it's more about managing the system, which is a very good example, then compile language is like a go-to. You don't have to think about it twice.
00:29:31
Speaker
And I completely agree with that.
00:29:36
Speaker
but good Just to add up in like one, not more than 30 seconds, on the carbon aware computing is very interesting because we are running these days, eblo we are running for ADEM study on carbon aware computing.
00:29:49
Speaker
And we are trying to modelize the effects of carbon aware computing in a consequential approach. Consequential meaning we are not just accounting for environmental impacts that would be spared or that would be generated. it's We try to modelize both the positive and negative effects of having carbon-aware approach. It's very interesting because it's it it shows, it's it's not released yet, but it shows that it's a very complex topic.
00:30:18
Speaker
and the The thing in ICT is that when we talk about carbon-aware computing, most of the time we talk about carbon-aware computing from a software perspective. And the thing is, there are multiple layers between the software and what actually happens on the electricity grid. And when we look at those constraints, constraints it's not so obvious that carbon-aware computing can save a lot of greenhouse gas emissions. It can in some situations. It seems like i will really brush a very...
00:30:49
Speaker
raw portrait of the study, but it it's more it seems more interesting on the time-shifting perspective, like shifting at moments when it's better, than geographical shifting. Geographical shifting seems very, very dependent on many, many variables, which makes it makes it not obvious as a saver. So yeah, I won't spoil the study that's not even finished, but it's a very, very complex topic that's very, it it would be a rabbit hole to discuss about that in detail.
00:31:20
Speaker
Brilliant. that' um That's really interesting. And especially that geographical versus time differential. Right. I think, again, it's it's often the the less obvious points that make the most sense in the end. And again, it proves that the the benefit of research is super important in making any decisions and talking to experts such as Benoit.
00:31:37
Speaker
Sometimes we're talking about both time and geographical because it's ah it allows you to run something like it was real time but and in another geography that benefits from having a night or being less consuming at this moment. So that's why yeah we try, we often make the assumption that both are related geographical and and and and time shift.
00:32:04
Speaker
and But yeah, it totally depends on the use case, I think. And and you you also have to take the whole ICT, not only rely on only the energy. for yeah Yeah, very, very interesting. and And talking about things that are sort of related and often bundled together, right, that maybe are actually quite separate. I mean, my next question is for you, Benoit. It's about green ops versus fin ops, right? and And the question is, if if we manage to decrease the cost of something,
00:32:32
Speaker
an instance, a compute, whatever, an environment, but we don't manage to reduce the energy and water consumption.
Balancing Financial and Environmental Impact
00:32:39
Speaker
Like what metric is more important and how do you decide? I mean, I think I know what some companies would say, but surely, and I think maybe there might be an opinion out there that actually if cost falls, water and carbon, you should fall too.
00:32:54
Speaker
But of course, that's not necessarily true. you know So in that very rough question, Benoit, what would you say? you know how would you answer that? Yeah, the Green Ops versus FinOps is a very interesting one as well. um Of course, as you mentioned, there is no one straight answer, very straight two points. But it seems like in some ways um you could benefit on both, on some actions.
00:33:23
Speaker
As Julien said, when it's actions that actually reduce the global capacity you are allocating or reserving for your project,
00:33:35
Speaker
In many cases, you could benefit both from a financial perspective and an environmental perspective. So that makes sense in many cases. That's pretty safe. um Now it's more about should I buy some spot instances on that cloud provider? Should I buy either reserve instances? you know That kind of optimizations that most of the time are motivated by money.
00:33:59
Speaker
um It seems like it's not so obvious. abuse It seems like there is there is a study from Teads, which is ah an advertising company in in France. It was a few years ago. It's a bit old, but still it was interesting. They showed some metrics from their platform. And they showed that while they managed to reduce over time the number of VCPUs allocated to their projects and that it allowed them to save money um in terms of because they continuously evaluated the environmental impacts of it, it was not decreasing. So it was an example of what what you say. Sometimes it's not just as simple as that.
00:34:44
Speaker
ah So, of course, there are many companies that will say it's okay. We're saving money, so we want it. Thing is is that if you if you don't have proof that it's reducing the environmental impacts, you might also reinvest this money in another project.
00:35:02
Speaker
and allocate new resources. So you might also, in a consequential approach, once again, you might also increase your impact by this way. So that's why it's always about what is the perimeter I'm looking at.
00:35:14
Speaker
i am i Am I looking on one system, or am I taking a step back and looking at the impact of what I do on the system them on the others? And then that's why the the the answer is not as simple as that.
00:35:28
Speaker
Yeah, brilliant answer. Brilliant answer, Benoit. I really like that. i think that's very interesting take on it. Julien, what do you think? Well, I totally agree. Pretty aligned on this. And I will just add that, yeah, it totally depends and so on the on the company, the policies and so on of the company. If they are publishing ESG commitments also, if they have some commitments about reducing their...
00:35:58
Speaker
the greenhouse emissions and so on. Investing money on the greener solution could also be used for communication, greenwashing between codes. so So, yeah, it could also minimize the lost but so loss the on the cost because if they're able to create some communication and having a brand that is more ah ah eco-friendly, I fucking had this term, but you hit the point. But the but yeah, it could be also investment saying, oh okay, even if it costs a bit more, but it's greener, then it's okay.
00:36:44
Speaker
it It also does really feel like it falls a little bit into this macro micro argument, right? Like if you're too focused on your micro on the specifics and you're not looking at that macro, you're not really understanding what's going on in the wider.
00:36:55
Speaker
And then, I mean, the phrase kept on repeating in my head, the butterfly effect, right? Every action we take has an equal and opposite reaction somewhere else. And when we're playing with carbon and the environment and and also finances, these butterfly effects can be, they can ripple and they can be really impactful. um Fantastic. um Okay. Um,
00:37:15
Speaker
Julian, we're going to go back to you um with this next question.
Using IDPs for Sustainable Development
00:37:19
Speaker
um So it is something quite key to kind of the cycloid world. It's, um you know, when we look at internal developer platforms, right, they're very good for structuring and and orchestration and also kind of baking in, you know, guardrails and things like this.
00:37:33
Speaker
How can we use IDPs, internal developer platforms, so that developers don't need to read policies, right? Because I do believe that human nature is probably one of the problems that we have in this industry, that people are not lazy, they're busy, right? And they're off on other tasks.
00:37:47
Speaker
How can we use structures, technologies like IDPs to help implement defaults so that developers don't have to read or other people or other end users, right? Can we do that? Is that a sensible way to do or are we really just...
00:38:01
Speaker
sweeping the problem under the carpet and leaving it for later, for the next CTO or for the next leader. Yeah, totally. I think IDP could be really great for that because you can remove some cognitive load from from developers and people who will use your service catalog. So if you have a a good policy on the on the platform team, we will create a whole service catalog and so on. You can, for example, defaults should be the use of a green region with lowest carbon intensity using data center with a better PUE that has been selected by the the platform team depending on the
00:38:44
Speaker
policies of the company so developers don't have to worry about how about this they're automatically using the the best uh the best approach uh so yeah you remove some cognitive load Your IDP could also provide some defaults about automatic shutdown for non-production environment, nightly, on the weekend.
00:39:10
Speaker
ah There's ah some studies and reports that mention it's almost 30% of the cloud spend that are wasted on this topic page only for stuff running during the night and weekend, for example. fall so so So yeah, you could create some sub-READ without having to to to to make the police ah to to do the bad cop with your employees and so on. So yeah, to finish, um I would say it depends on the on the container image also and application runtime. Lighter code, lighter containers will have a lighter footprint. So if you enforce the standardization across the whole company, to to use the best
00:40:00
Speaker
um as a the best code approach and an infrastructure approach for reducing emission could be a could be it could be really nice. so So yeah, that's why I think IEDP could really help on this part.
00:40:17
Speaker
Fantastic. Thank you. Yeah, I think it's really important. And of course, yeah human nature always comes into that. right And of course, if people are, ah you know, made to use the right settings and and forced to sort of enact their work within a very specific framework. It's good for them.
00:40:34
Speaker
Benoit, what do you think? I mean, are there other technologies out there? Are there other ways that we can bake in defaults um you know as ah as a kind of consultant advisory role? How would you advise a company to implement things like this if they're you know yeah just if they're trying to do this, basically? I won't be rude about developers who don't read policies.
00:40:52
Speaker
Yeah, I think about at least two ways. One would work with IDP and would make sense with IDP. I think it's like developing under constraints. Like ah we could imagine an IDP that gives, you know, default sizes for, I don't know, instances for...
00:41:12
Speaker
database capacity for any technical component and say, hey, on that kind of project you are working on, regarding the other project we have made in the past, you should not need more than this kind of capacity for a simple setup.
00:41:29
Speaker
And so if you, we could imagine an IDP that says, i you you you are close to the max capacity, so maybe you could optimize something. And maybe you could go to that person who managed this project before you, maybe he might have some insights. So that would be one way, developing under constraint. I think I really believe in this.
00:41:49
Speaker
And the second one is, even if it's not the default go-to, but still it can be useful in several situations, it's giving continuous insights on what you are going, so more different kind of metrics. So I realized something working on Skaffone and on other projects after that, I realized that Actually, if someone wants to optimize its code or system, it doesn't need energy metrics so much.
00:42:21
Speaker
Actually, you could just focus on memory consumption. You can focus on how much data you send the network, on how much data you store on storage, how much compute capacity you are actually using with your app. ah It's just that we we are lazy. So having an ener energy metric shows you that, hey, it has an impact on the real world. It's not just IT resources that kind of seem virtual because we are in the cloud most of the time. It has an impact on the real world. So it's just a translation. explain it ah as a translation. um
00:42:58
Speaker
So even if that's for sometimes wrong reasons, adding some metrics that have a meaning in the physical world, kind of works as well.
00:43:10
Speaker
You could add, and what I encourage you to do is not going just through energy matrix, going through environmental matrix. So estimation of both embodied and operational emissions, adding matrix on maybe water consumption on maybe biotic resources depletions which means minerals and metals so it might not work for everyone but it will ring a bell to many people let's say how is that come that i am consuming virtual instances and then it consumes water what does that come from then you can go to the documentation then oh okay it doesn't impact it might raise awareness i think in that way that's interesting
00:43:48
Speaker
I really love that. Only the metric would be enough. Something we we plan to do at Cycloid. It's on the roadmap. I don't know exactly when it will be released. ah But we'd like to to gamify with some scorecards and so on, those metrics. Because if you if you check only the metric and awareness, I think we have a really good example with ah smoking smoke kills.
00:44:16
Speaker
ah uh people were smoking ah we thought we told them you will have a cancer you will have a disease you will have and people were still smoking and finally we had to take some action to forbid people smoking and this really worked this uh with we ah forcing people we observed uh decreased and if you try to just inform ah Of course it will work for some people, but it will work for people who are already informed and people who are willing to do to do things better. But for people who are not really aware about it and not really concerned, ah finally fon it won't, maybe won't be enough. but the brain has some pieces and some and we we like game and we like perfection we like scores we like to to have a good ah result a good appreciation saying oh you're the best guy so maybe trying to gamify this in order to improve this uh you have your your dopamine uh shot on your brain saying okay i did something great even if you don't really know exactly what you did
00:45:25
Speaker
But if you have ah something giving you a thumb up, say, oh, yeah, I'm really good and I'm really glad to. i did my week ah sprint planning, and that's that's great.
00:45:37
Speaker
so So yeah, I think it's really important. But currently, what we're we're doing at Cycloid, we're showing for every developers when they're working on the project, we're showing the cost. We're showing also the carbon footprint.
00:45:52
Speaker
of the project. They're able to know if it's increased, if it decreased, depending on the releases they've done. ah But yeah, trying to having some scorecard and trying to gamify the process, I think will have a even a better impact, a bigger impact than than just having the metering.
00:46:13
Speaker
I think that's brilliant. Translation, dopamine, and management with a whip. I think this is exactly the way to go on this one. I think it's it's really good. Gamification works supremely well. And yet we are all slaves to the dopamine hits, right? And this idea of translating into reality and translating into di direct action, I think is is a really nice concept. I've never heard that word used in this way to sort of describe this this kind of concept.
00:46:34
Speaker
um Okay, we'll do the the final last long question now. It's for you, Benoit. And it's ah it's that one that we've been waiting
Critique of PUE and WUE Metrics
00:46:42
Speaker
so long for. It's PUE versus WUE. So it's power, water usage, effectiveness, right?
00:46:48
Speaker
ah The question is, Are they really useful KPIs or are they quite easy to game, to cheat on, to ah to present um in a shiny way when they're not actually really, and but they're more kind of darker than they appear?
00:47:01
Speaker
are they Are they fit for purpose in 2025 or are they actually fit for cheating on in 2025?
00:47:08
Speaker
i Yeah, i I will start with you the with the dark side of it. So the dark side, let's start with PUE. is that first, it's an efficiency metric.
00:47:22
Speaker
So as always, efficiency means optimization at best, doesn't mean reducing the impact. um We had a discussion with Julien before, and he said very rightfully, he said, you can increase your energy consumption for IT and decrease your PUE. That's how it works. It's just a fraction.
00:47:43
Speaker
With IT energy as the as the divider. So you can just, yeah. So having a better PUE, it's very funny because every time you browse about it and sustainability, you will hear about PUE. Everyone would say, yeah, but this provider has lower PUE. So they are very optimizing stuff. And so, yeah, they probably optimizing stuff. Doesn't mean they have a lower impact than the other data center.
00:48:10
Speaker
Not at all. It depends on the size of the data center, it depends on how much energy you consume as a whole and not just a fraction. ah And there are many, many aspects where the PoE is gamed, as you said.
00:48:29
Speaker
Sometimes it's consciously, sometimes it's not very it's not an intent to game it, just the metric has flows. First, it's always represented as an average.
00:48:40
Speaker
So it's a yearly average most of the time. But PUE depends on how how important is your load. your IT load. It depends on the seasonality, depends on the outside temperature because you will consume more or less energy for cooling depending on how hot it is outside. And so you don't see that. You you just focus on one metric that is fixed.
00:49:05
Speaker
That sometimes is not even measured in real conditions. It can be measured when the data center has just been built with some machines to simulate ah working situation, putting the machines at full capacity, and then you measure the PVE.
00:49:22
Speaker
So you can have much lower PVE this way than what you would have if you measure in actual working situation. So, yeah, so that's the dark side of it.
00:49:37
Speaker
There are norms that try to standardize the way we measure and we display the PUE to make it more comparable or use useful. Most of the time, it's not respected, even by the biggest providers. the you You will see it very easily. easily The few providers that respect it, they provide audit report on their website.
00:50:00
Speaker
i was It's not too... say that they are the best. That's not the the point. It's just on this very specific point, one of the few is OVH. OVH, on their website, they have the audit report for how is measured the PUE, WUE. That doesn't mean they do everything good on those aspects. Just they are transparent on this.
00:50:20
Speaker
And there is another company that assessed how they measured it. That should be the norm. And that is respecting the norm, actually. But most of the time, it's not the case. On the bright side,
00:50:33
Speaker
can be useful in a relative perspective. It's still something that even if we know it's flawed, even if we know it's not exact because not displayed the right way, for especially for research, that's that's interesting to still have it because we can see trends, we can see who is obviously gaming it, who is actually somewhat in the average of the industry, so maybe it's possible. We can see it helps, it gives insights. um But the thing is, maybe the... So to to wrap it up, maybe the biggest problem of PUE is not having PUE and using PUE. It might be useful as every metric. The real problem with PUE is that it's a totem.
00:51:21
Speaker
It's like talking about sustainability to a data center or cloud provider, they will say, oh, PUE. So I kind of have a PUE fatigue because it doesn't mean anything in a sustainability perspective. um And it doesn't mean it's not useful, but start to show real metrics and retrospectively, like properly filling your energy efficiency directive report, which is now becoming a law in all EU members states. ah We know that the databases are still very, very empty while... So yeah, there is no...
00:52:01
Speaker
there is no um financial incentive to do it. So that's why. But why should we wait for financial incentive? If you are really transparent and you really want to improve the topic, just fill your report in.
00:52:14
Speaker
That would be your first step. And there you would fill very interesting metrics about sustainability of your data centers, not just PUE. Brilliant. What a great answer. And yeah, what a totem to avoid.
00:52:27
Speaker
Julien. Any thoughts? I know you talked a but little bit about it before, but I mean, of course, this is, again, something very close to cycloid in your own heart. Like, you know, yeah, are these gamified metrics or as Benoit says, or, you know, can we actually try to see that that that light and that brightness through them as well?
00:52:43
Speaker
Yeah, but that's not really my expertise field compared to Benoit, which is ah really ah its core expertise. um but on On my side, I would say for everyone who will ah listen to this podcast, I think PUE is like the gross domestic product. It's ah something we The formula we invented at some point because there was nothing in order to to to calculate something and we we knew it wasn't perfect. We knew it must be improved and must be thought deeper and better. And the thing is that we're using we're still using the same formula since 20 years, even more for the closed domestic product. And we and finally, it's totally
00:53:34
Speaker
Well, it's just a formula that gives a result, but the result as a is meaningless, finally, because it doesn't compare things correctly and so on. And same for the cross-domestic product. we We should have improved it since so long. Even the creator, the author of the cross-domestic product, said it's not a perfect thing. and It must be not be used.
00:53:56
Speaker
And we we're still using it 70 years ah later. And same for the PUE, and I hope we won't use it for 70 years or more. So yeah, that's my point. i think ah I think it's ah it's not really relevant when everyone aren't playing the same game.
00:54:18
Speaker
ah so So yeah, difficult to compare and difficult to to really trust this number.
00:54:29
Speaker
GDP versus UE. Both of them nonsensical. No, I think it it really it really makes sense that, you know, in many ways we've've we've found a metric, we stuck with that metric, we've built systems around that metric, we've built understanding around that metric, and people just are not willing to leave that metric in the past where it belongs.
00:54:47
Speaker
um you know And it it sort of, it shines a light in a way, sort of almost the same thing on shadow IT, when we look at translating understanding into a real rational reasoning. when we look at optimizing versus actual action you know when we take all of the things that we've talked about it's that we have the path to follow we are able to go that way but we seem trapped a little bit in our past and a little bit unable to be perfectly truthful to use one word truthful right with ourselves we know that a lot of the metrics that we're using aren't really not they're not fit for purpose we're not just not using them correctly so
00:55:21
Speaker
I think for me, this has been incredibly interesting, Benoit and Julian. And I just want to say thank you both very, very much um for the time that you've given us today. and Really interesting chat, lots of learnings, lots of facts, many new perspectives on a green IT on a cloud.
00:55:37
Speaker
And yeah, I just want to say thank you very much. um Julian, thank you. Benoit, thank you. if You just want to say any final words to our audience? Julian, I'll let you go first. um No, thank you for the invitation. i Really, are really happy to to talk about those topics. And it's always a pleasure to discuss with Benoit. So we'll meet ah soon also on the Hublot's day. So yeah, always a pleasure to to discuss about it.
00:56:05
Speaker
And Benoit? Yeah, thank thank you for hosting us. And it was a pleasure. So yeah, i for for for the presentation, The listeners, I would just encourage them to go fetch some new resources. There are plenty of them, whether it's Boa Vista or other NGOs who are doing very great work. There are many, many of them. The documentation is here. So actually, if you really want to better understand the topic, it's out there. It's really easily findable.
00:56:33
Speaker
So just go for it. And even if you don't start with the most impactful measure first, that's not very important. you What's very important is to start with something and to start really trying something, not just saying, oh, yeah, will do it in six months. In six months, it will be six months of too much greenhouse gas emissions emitted already. So start now.
00:56:54
Speaker
Yeah, brilliant. And that is a great note to end on. Start now. Brilliant. Thank you very much. And just like the Exfeld and Mulder and Scully said, the truth is out there if you search for it.