4:02

So, Kim, when Kara says ESG,

4:05

what does she mean? ESG

4:08

stands for environmental, social,

4:10

and governance. Okay. And it first came to prominence

4:13

in reporting from the United Nations. It's

4:15

a framework that organizations can use

4:17

in their own corporate strategy to deliver

4:20

a value for all scenario

4:22

as it pertains to areas like pollution

4:24

or environmental conservation. But

4:26

it's not just for the environment. It also

4:29

addresses working conditions for employees,

4:31

privacy practices for customers.

4:34

So companies using

4:36

the ESG framework depend

4:39

on real-time data. Projects

4:42

like OS Climate create a way for

4:45

that data to be centralized on

4:47

a common platform so that data

4:49

scientists can build a model to show a clear

4:52

picture of, for example, a company's

4:54

energy consumption, their carbon

4:56

footprint, and other things. This

4:58

is a topic that I am not familiar

5:01

with at all. Sustainable

5:03

finance? It sounds like a really

5:05

weird made-up term when

5:07

I think finance, I think money. But

5:10

I would love for my finances to be sustainable.

5:13

I think we all do. So you

5:16

have my attention, but when you broke it down

5:18

to what it really means, being

5:20

good citizens and lending money for social

5:22

justice and green projects, that is huge.

5:26

When we're talking about mass corporations,

5:29

I always thought that that's where the

5:31

change should happen

5:33

at the corporate level, because

5:35

when sustainability happens, it

5:37

should really trickle down to the rest of us

5:40

and not the other way around. Let

5:42

the corporations make the sacrifices

5:44

and show how it should be done. We

5:47

can see things happening, and then

5:49

we can trickle down and say, this is

5:51

how sustainability should work. So

5:53

I know this is a new term for me, sustainable

5:56

finance, but I really like what it stands

5:58

for.

5:59

Well, I think this framework of

6:02

ESG, it's something

6:04

that I hear a lot about in corporate strategy.

6:07

This is something that a lot of companies are talking about.

6:10

What I didn't realize and what

6:12

I'm learning from Kara is

6:14

that

6:15

a

6:16

lot of the strategy relies on data,

6:19

right?

6:19

Yes. Which is

6:22

a technical problem. And

6:24

who better to address this problem

6:28

than someone like Kara, right? Right.

6:32

So you're probably wondering though, this

6:34

is a lot of talk about sustainable

6:37

finance, assets, risk models.

6:39

What does that have to do with data centers? Along

6:42

with all the work she's doing with OS Climate,

6:45

Kara is also a technical lead within CNCF,

6:48

the Cloud Native Computing Foundation. She

6:51

represents Red Hat on the CNCF's

6:53

technical advisory group. It's kind

6:55

of a cross section of tech companies. Some

6:57

of them are small and some of them are the biggest

6:59

names in the industry. We are

7:02

working

7:02

together to have combined efforts

7:05

around carbon tracking, creating

7:07

best practices for carbon

7:09

tracking for, especially around data

7:11

centers and anything around cloud

7:14

native technology. We actually

7:16

do have a project that

7:19

has been sent to the CNCF,

7:22

Project Kepler, that is intended

7:24

to help with some cloud

7:27

native technology and data

7:29

centers as well.

7:31

It sounds like there are a lot of parties involved

7:33

in trying to figure out this problem

7:36

and trying to figure out data centers' impact

7:38

on sustainability. But what exactly

7:41

is Project Kepler and what does it

7:43

do? We'll tackle that next.

7:58

Is

8:00

it? That's a mouthful. It

8:03

is. It is. Well,

8:05

what is this, Kim? Yeah, I don't

8:07

know. OK.

8:09

I really don't. I

8:11

thought I think I know. I let's

8:13

see if I have a handle on it. OK. But

8:15

let's get Kara in real quick to give us

8:17

some background.

8:18

OK. OK.

8:21

Kepler comes into the story from

8:24

an IT optimization standpoint,

8:26

so it's optimizing energy

8:29

and tense processes and really

8:31

making supply chains more transparent.

8:34

It was founded by Red Hat and

8:36

within our Emerging Technologies Group, which

8:39

is actually where I sit. And

8:41

it was created in collaboration

8:43

with IBM Research. It

8:45

was intended to capture power

8:48

usage metrics from Kubernetes

8:50

clusters to see where their

8:52

efficiencies to be more effective.

8:54

And it uses EBPF,

8:57

which is Extended Berkeley Packet Filter

9:00

in the Linux kernel to be able

9:02

to use machine learning models

9:04

to estimate power consumption by the

9:06

workload in a way for

9:09

it to be able to be exported as a set

9:12

of Prometheus metrics for tracking

9:14

carbon footprint.

9:16

All right. Angela, can can

9:18

you help us decipher

9:20

some of this? Brent. Is

9:24

that a yes?

9:25

OK, let me try my hand at

9:27

this. OK. I may

9:29

be way off, but what

9:32

this project is about is about tapping

9:34

into energy consumption in Kubernetes.

9:38

So there has to be this tool

9:40

and it's seeing, well, how much

9:43

energy, how much power is being consumed

9:45

by all these different processes inside

9:48

the cluster, different containers, different

9:50

services, and it's trying to figure out the

9:52

metrics. Where is all this consumption

9:55

coming from in the cluster? And

9:58

with this tool, EBPF.

9:59

and yes, I've heard of it and I've

10:02

never used it, but especially in this context,

10:04

if this is a tool you use to figure out

10:07

how much power is being consumed with certain

10:09

workloads and then you're able to use,

10:11

say Prometheus, which is great

10:14

to visualize your metrics, then

10:16

you get an understanding of how, you know, some

10:18

of your hungriest workloads may behave.

10:21

And then you can figure out, well, how do I offset

10:24

some of these very consumptive workloads?

10:27

So it sounds as if we're trying to figure out a way

10:29

to look at technology a little bit smarter

10:32

and how it's consuming energy and maybe

10:35

having a better way to track it, question

10:37

mark.

10:37

And I think optimize, right?

10:40

Yeah, exactly. So she said optimizing

10:42

for energy intense processes.

10:46

Yes, and the optimization part will come in

10:48

a little bit later. But that

10:50

was pretty spot on, Angela. These

10:53

metrics are then, you know, these

10:55

Prometheus metrics are then used for scheduling,

10:57

scaling, reporting and visualization, which provides

11:00

sys admins with the information on

11:03

the carbon footprint of their workloads.

11:05

So performance metrics, tracking

11:07

performance metrics platforms for doing so, that's

11:09

nothing new, right? But you

11:12

have to keep in mind, a lot of these sources

11:14

of data are disparate.

11:16

They can often be walled behind

11:18

proprietary, you know, proprietary

11:20

software. They're not housed in the same

11:23

place, but Kepler

11:25

can change all that. That

11:28

transparency where around

11:30

the metrics that some other

11:32

metrics providers, you definitely, because

11:34

they're not open source, you're not able to

11:36

see all of the sources or all

11:39

of the inputs into those metrics.

11:41

And so Kepler continually

11:43

adjusts and fine tunes

11:46

through pre-trained models using

11:48

node data from power

11:51

estimating agents that are running on

11:53

servers. And so those metrics

11:55

can be combined with power

11:58

usage to calculate.

11:59

the carbon footprint of that

12:02

workload. So if you're

12:04

looking at the workload, you can see where the

12:07

opportunities are, whether

12:09

or not you power down a

12:12

certain workload or you power up one,

12:14

they can, based

12:17

on what are the outputs of the carbon

12:19

intensity. You called it, Brent. Spot

12:22

on. Optimization. Optimization,

12:24

right? Yes. So let's

12:27

take a second and think about, when you think

12:29

of a data center, what is

12:32

it? We're talking about, obviously, a lot of

12:34

racks, right? A lot of servers, a lot of

12:36

racks. We're talking about AC

12:38

running at 40 degrees Fahrenheit,

12:41

24 hours a day, seven days a week, 365 days a year.

12:44

That's kind of how I imagine it. Am I off?

12:47

You are spot on. Everything

12:49

is always running all the time inside

12:52

of a data center. Every server, every

12:54

piece of equipment, everything at the top

12:56

of the rack requires energy and it's

12:58

running constantly. And it's

13:00

doing so because you don't

13:02

want your circuit board to overheat, you don't want

13:04

your equipment to fry. So that's a lot

13:07

of air conditioning. And if you think about

13:09

it, when you're home, you cannot imagine running

13:11

your air conditioner all year round, correct?

13:16

What's so wild about this is that, the

13:18

way we talk about data centers and especially

13:21

the cloud, it's such an abstract

13:24

concept.

13:25

But is it

13:27

though? No, it's

13:29

not. I think that's- Okay, just think about this

13:31

for a minute. The data centers that we used to house

13:33

in our buildings, in our offices, they've

13:36

just moved into bigger buildings. So

13:39

it's like everyone's data center just up

13:41

and move somewhere. And all of that consumption

13:43

is now happening in one place.

13:45

What's really interesting about

13:47

all this is like, the way that we typically

13:49

talk about data centers, right? We

13:52

talk about the cloud. It's

13:54

very ephemeral, right? It

13:57

feels very abstract, but

13:59

it- They're very physical

14:01

objects that need a very particular

14:04

environment. And from

14:06

what I hear you saying, Kim,

14:09

they also use a lot of energy.

14:12

Yes. Some kind

14:14

of numbers we have, which obviously

14:17

are, you know, dated

14:19

at this point, if you're thinking about how

14:21

many data centers there are in the world and how

14:24

this technology is proliferating, it's

14:26

increasing 10 to 30 percent over a year

14:29

because, you know, you're getting more and

14:31

more, more and more data centers, more and more workloads.

14:34

It's just a very kind of exponential type of

14:37

growth. According to the International Energy

14:39

Agency, data centers account for

14:41

about 1.5 percent of

14:44

all global energy consumption. And

14:47

if you go to just the United States, where

14:49

I reside, it's 3 percent of

14:51

electric

14:52

power. Let's

14:55

pause here for a second because those are some

14:58

kind of wild numbers.

15:00

Yeah. They don't seem like a lot, though, right?

15:03

Like 3 percent is like, all right. You know what I mean? But

15:05

if you think about like the total, like

15:08

total electric power in the United

15:10

States, 3 percent of I don't

15:13

even know what that is, but I'm guessing 3 percent

15:16

of that is like a lot

15:18

of energy.

15:20

I would probably say it's enough to power

15:22

a couple of cities, like

15:24

mid-sized cities. That may

15:26

not seem like a lot, though. It still doesn't like 3 percent.

15:29

It's like compared to, you know,

15:32

other things that we use every day. Like if you're

15:34

talking about consumer electronics, it doesn't

15:36

seem like a lot. But with the needs increasing

15:39

over time, like I said, 10 to 30 percent year

15:41

on year, ignoring it isn't

15:44

exactly a good

15:44

plan. Kara

15:47

says that Kepler can integrate

15:49

with Kubernetes to not only display

15:51

workloads and their energy consumption, but

15:53

again, it can optimize their

15:56

performance. The Kepler metrics

15:58

can be employed by a.

15:59

Kubernetes scheduler

16:02

to place the upcoming workload

16:05

on the compute node and it

16:07

can improve performance per watt

16:10

and again it's through auto scaling

16:12

algorithms and so the

16:16

nice one of the nice parts too is

16:18

that it can be integrated with continuous integration

16:21

and deliveries OCICD

16:25

pipelines to optimize the

16:27

efficiency of energy consumption and

16:30

they can actually be placed into some sort

16:33

of dashboard that really presents

16:36

what the power consumption is at different

16:38

levels and include containers,

16:41

pods, namespaces or different

16:43

compute nodes in that cluster. That

16:47

is pretty interesting because I'm trying

16:49

to figure out

16:51

when the scheduler decides

16:53

oh where am I going to put this workload right?

16:56

Is it looking at the most optimal

16:59

hardware in which to place

17:01

these workloads? I mean are we talking about

17:04

servers that have a more greener

17:08

for lack of a better word you

17:10

know workload that can hold those workloads?

17:13

I'm really trying to understand like what

17:15

does the scheduler it says okay I have this

17:17

workload and I need to scale it and

17:19

I need to make sure that it is

17:21

definitely on its most efficient

17:24

systems. Yeah and I'm wondering if it has

17:26

a lot to do with how the hardware

17:28

behaves and if we're

17:30

and that's a good question so I

17:33

know that it has a lot to do with

17:35

how the energy they're tracking how

17:38

energy is consumed but the

17:40

hardware has to play a part that's all I'm saying.

17:42

Yes absolutely. Like what where

17:44

does the hardware come in and how do we decide

17:47

you know the

17:49

most efficient hardware to

17:51

place said workloads on? My response

17:53

to that would be that would depend

17:57

on the infrastructure and

17:59

the

17:59

at the company that chooses

18:02

to adopt Kepler, right? Like

18:05

they would be kind of like the, I guess, owners

18:08

of the hardware question that you're posing,

18:10

Angela. But it's very interesting. And honestly, I didn't

18:12

really think about that before, but you do have a point.

18:15

Because not all hardware is created equal. There

18:18

may be a question of which server

18:21

is more optimal and which one

18:23

runs, I guess, quote unquote, greener than the others. So

18:25

a company, let's say a healthcare

18:27

company, is

18:29

trying to get a sense of their energy usage. Maybe

18:32

they think the spikes in their energy consumption

18:35

are related to when doctors are accessing health

18:37

data or when they are, for

18:39

example, a very busy hospital is admitting

18:41

a high volume of patients on certain days of the

18:43

week. The company or

18:46

the organization that runs the hospital could

18:48

use a project like Kepler to build out their

18:50

own comprehensive dashboard to

18:53

observe these patterns of energy consumption and

18:55

verify that information instead of having it

18:57

just be, I don't know, I guess. They

19:00

could even use the machine learning features within

19:02

Kepler to estimate future

19:04

usage and then modify

19:06

whatever internal infrastructure that they need

19:09

to modify. Huh.

19:11

Interesting. Does that sound right? Oh,

19:13

for sure. It sounds like we're moving in the

19:15

right direction. I mean, but it still all

19:17

boils down to the hardware, right? That's

19:20

always going to be the final arbiter as to

19:23

how efficient these workloads are running.

19:26

To be able to support a higher volume

19:28

of patients, you know, you're going to be

19:30

scaling up usually. And that

19:32

means more pods, that means more containers, that

19:35

means more people accessing

19:37

that endpoint. And that

19:39

is an energy consumption

19:41

increase. So how is

19:43

your hardware going to handle this? And

19:45

being able to see those peaks

19:47

and valleys, maybe you can plan

19:50

better. Maybe you can do,

19:52

when you do your hardware procurement, you can kind

19:55

of move in that direction and

19:57

see how those workloads kind of ebb

19:59

and fall. flow and then when you bring in more energy

20:02

efficient hardware, you can see

20:04

those same peaks and valleys, but maybe

20:07

they're not peaking like they used to

20:09

because you're taking an alt to the consideration

20:12

how Kepler helps you

20:14

see the efficiency and then Prometheus,

20:18

where you see those metrics and you're looking at that

20:20

dashboard and you can react and

20:23

interact with those usage spikes.

20:26

Right. Does that sound... I

20:28

feel you on that though. That

20:31

was a great story or use case to

20:33

make it make sense.

20:36

I guess a good spin off from that

20:38

would be the difference between

20:40

an on-prem situation or an off-prem

20:42

situation. If

20:45

you're dealing with a data

20:47

center that a, for example, I'll use

20:49

my hospital example again, they don't necessarily

20:51

control that data center and

20:53

they may not control the hardware that's being

20:56

used. Maybe there's

20:58

something on a macro level where

21:01

parties that control the data center can

21:03

be incentivized to use

21:05

greener servers or to use servers

21:07

that can be optimized

21:09

to run more efficiently. It's definitely moving

21:12

in the right direction. If we're trying to be more

21:14

energy conscious and we're trying

21:17

to be more green as a community,

21:20

as a bunch of technologists, the technology

21:22

community, and we're building projects

21:25

like Kepler and we're graphing

21:27

them and we have sustainability

21:30

finance where we're trying to make sure that

21:32

we're doing the right thing with our

21:35

resources, talking about the greater good,

21:37

then we should be behooved

21:40

to make sure that what can we do

21:42

to

21:43

be better? Do we have to

21:45

buy different hardware? Do we maybe

21:48

move from one cloud to a more

21:50

energy efficient cloud if that's a thing? I

21:53

don't know, that may become a thing if

21:55

you think about it because if you have taken

21:57

up the mantle as an organization

21:59

that can... cares about sustainability, you

22:02

may be looking for a place to run your

22:04

workloads that believes the same thing.

22:07

And they have things in place to make that

22:09

work. Wow.

22:10

Angela, you just said, like, what can we do?

22:14

And as technologists,

22:16

what I'm hearing from Kara is that there's

22:18

quite a lot that we can

22:21

do.

22:22

Indeed. It's always in our hands.

22:25

You know, it starts from up top, but then when we see

22:27

that, how changes are being made,

22:29

well, we can start making those changes, too. Maybe

22:32

it can mean how efficiently

22:34

our code is running. You know, how are we

22:36

looking at, you know, what our code

22:38

is doing? Is it the most efficient

22:41

way to run a particular

22:43

process? Are there more efficient ways

22:45

that take up less resources? Yes. Do

22:48

we tweak our programs to make them more

22:50

energy efficient? And I don't even know

22:52

if that's a thing. I'm just going to assume it

22:54

is because, you know, everything's possible

22:56

nowadays. But I think

22:58

there's something we can do. Yeah. We're

23:01

talking about being more sustainable. Yes. And

23:04

then I would think in like an open source community

23:06

kind of collaborative spirit, I

23:08

feel like there's a space

23:11

for technologists to kind of reach

23:13

beyond their teams and even beyond their

23:16

organizations to kind of advocate for these cross-sections

23:19

of solutions for hardware and software.

23:22

I

23:22

think that there's a lot of room

23:24

and maybe even a lot of appetite for that

23:26

kind of collaboration, that kind of collaborative effort

23:29

to offset or

23:31

at the very least reverse what's

23:33

happening with energy consumption for data

23:35

centers.

23:35

And this is like sustainability

23:38

and climate change and things like that. It's

23:40

one of those big problems that like no

23:42

single person or no single

23:44

company can solve on their own. Like

23:47

it takes a lot of companies

23:50

and a lot of people coming together to solve

23:53

a common challenge. Indeed. If

23:56

we're all working separately

23:57

in our own little silos or or

24:00

little companies or rooms

24:02

or whatever, we're all less

24:05

effective than we are working together.

24:07

Right, I think that's exactly what Kara

24:10

is also getting at. I

24:13

asked her as someone who found themselves

24:15

working in this space in a very unconventional

24:18

way, how she feels about

24:20

the work that she's doing now. And she says

24:22

kind of what you're saying, Brent, that it's

24:25

just more evidence that participating in

24:27

an open source community and collaborating

24:30

with different companies and different organizations

24:32

can push everyone towards a common

24:34

goal.

24:37

Whether they consider themselves

24:39

a technical or non-technical person,

24:42

anyone can take part in open source

24:45

and truly feel like open

24:47

source collaboration is where it's

24:50

going to help us to achieve climate-related

24:52

goals so much faster by

24:54

using a common set

24:56

of core practices and technologies

25:00

that are accessible by all.

25:02

I

25:05

want to

25:06

come back to something you said at

25:08

the top of the episode, Kim. You

25:11

were describing this, I guess,

25:14

internal conflict that you

25:16

had about something that you value, which

25:19

is sustainability, and maybe

25:21

how your profession or

25:26

the industry that you're in is

25:29

somehow contributing to that.

25:30

Yeah, that's interesting. And yeah,

25:33

there is a level, and I'm

25:35

sure I'm not the first person to

25:37

invent this feeling, but I

25:40

really care about the planet,

25:43

because I live on it, you know? We all kind of do,

25:45

and we all have to live here together. But

25:48

I've never honestly thought about sustainability

25:50

in this light. Technologists

25:53

have worked together to address so many different

25:55

challenges on a global scale, but it

25:57

makes sense that the lens would... also

26:00

be directed inward towards the proliferation

26:03

of data centers and towards the impact they have

26:05

on the power grid. I hope

26:07

that projects like Project

26:09

Kepler can influence how

26:12

we think about cloud native technology as

26:15

it evolves and it changes. And

26:17

I'm really glad to see open source communities

26:19

being a part of that and they're really excited

26:21

and galvanized to make change. Yeah.

26:24

And I agree, this episode was such

26:26

a clinic in understanding

26:28

a lot of new terminology

26:29

and technology. And

26:32

to really put it in perspective that we

26:35

play a part in protecting

26:38

our planet. And it doesn't seem like,

26:40

well, you know, where does our part come

26:42

in? But collaboration

26:45

and working in open source and

26:47

working together to solve these types

26:50

of problems. That's how it all,

26:52

that's how it works. That's

26:54

how the good stuff comes, you know, and we've seen it

26:56

time and time again. Open source

26:59

communities are where good things

27:01

happen when folks come together. So

27:03

I am very encouraged by projects

27:07

like Kepler and others that are

27:09

probably on the horizon that are addressing

27:12

climate change as well. And I

27:14

can't wait to see what's next because we

27:16

have to live here. So let's keep

27:19

it around for a little while longer, shall we? Let's

27:21

do what we can.

27:25

So what do you think about what you just heard?

27:28

Sustainable finance, ESG framework,

27:30

emerging technology, climate

27:33

change, and it's all happening within

27:35

the data centers and the power consumption.

27:38

There was so much information in this episode.

27:41

We want to hear what you thought. I know

27:43

you have thoughts on it because I do. Use

27:45

the hashtag Compiler Podcast. We

27:47

would love to hear what you thought about this episode.

27:50

Can't wait to hear what you got to say.

27:56

And that does it for this episode of Compiler.

28:00

The episode was produced by Kim Wong

28:02

and Caroline Craighead. A big

28:04

thank you to our guests, Cara Delia.

28:07

Victoria Lawton empowers and

28:09

sustains us every single

28:12

day.

28:12

Our audio engineer is Robin

28:15

Edgar. Special thanks to Sean Cole.

28:17

Our theme song was composed by Mary Anchetta.

28:20

Our audio team includes Lee

28:22

Day, Stephanie Wonderlick, Mike Esser,

28:25

Nick Burns, Aaron Williamson, Karen

28:27

King, Jared Oates, Rachel Ertel,

28:30

Devin Pope, Matthias Foundez, Mike

28:32

Compton, Ocean Matthews, and Alex Trebulzi.

28:35

If you like today's episode,

28:37

please follow the show, rate the show,

28:39

leave a review, share it with someone you know.

28:42

It really helps the show and we like to hear

28:44

about it.

28:45

All right, everyone. We'll see you next time.

28:47

See ya. All right.