Your cloud provider runs efficient data centers, but your workload design is on you
Sustainability takes decades while digital transformation takes months, yet reaching net zero means speeding up today. Katja Philipp, a Solutions Architect at Amazon Web Services who helps power and utilities customers on their cloud journey, explains why efficient data centers are only part of the carbon story. So which metrics and design choices let engineering teams cut the footprint of the workloads they own?
Learn the proxy metrics and five design areas that make cloud workloads more resource efficient
Katja Philipp, a Solutions Architect at Amazon Web Services, explains how cloud architects can treat sustainability as a design requirement alongside cost and security. She introduces the shared responsibility model for sustainability, proxy metrics that make carbon reduction measurable, and the five best practice areas of the AWS Well-Architected sustainability pillar, illustrated with a machine learning example.
“So the really important challenge here is with sustainability is how can we achieve an outcome that is decades away while we actually know in order to get there, we need to accelerate the process today.”
“And this is now such an important step to seeing sustainability as the same non-functional requirement as cost or security.”
What you will learn
Distinguish the cloud provider's responsibility for sustainability of the cloud from your own responsibility for sustainability in the cloud.
Define proxy metrics, such as resources per unit of work, that make carbon reduction tangible for application teams.
Apply the five best practice areas of the sustainability pillar to review an existing workload.
Reduce a workload's data footprint with lifecycle policies, colder storage tiers, retention periods and compression.
Plan an improvement cycle: set a sustainability goal, measure, prioritize changes and reassess the impact.
Key takeaways from the speaker
Shared Responsibility Model for Sustainability
Best Practices around Architecting for Sustainability
What does it mean when we talk about Sustainability within the IT and Cloud context
Details
Level
Intermediate
Format
Talk
Technical depth
Practitioner
For
Early career, Mid-career, Senior individual contributor
Length
22:45
Published
July 2022
Questions this talk answers
What is the shared responsibility model for sustainability in the cloud?
AWS is responsible for sustainability of the cloud, meaning data centers, global infrastructure and renewable energy. Customers are responsible for sustainability in the cloud, meaning how they design workloads, use data and write software.
How can teams measure sustainability when a total carbon figure is hard to act on?
By using proxy metrics such as compute, storage and data transfer usage, normalized per unit of work like vCPU hours per hundred orders, plus utilization per resource. These can be compared month to month despite fluctuating usage.
Why can migrating to the cloud lower a carbon footprint?
Centralized cloud data centers are optimized for energy efficiency and run at higher utilization than many small on-premises data centers. As the speaker cited, a benchmark found AWS infrastructure in Europe up to five times more energy efficient, with a possible footprint reduction of nearly 80%.
Which areas does the Well-Architected sustainability pillar cover?
Five areas: user behavior patterns, software and architecture patterns, hardware patterns, data patterns, and development and deployment processes.
How do data best practices apply to a machine learning workload?
Reuse publicly available data sets to avoid duplication, classify data and move it to colder storage tiers, set retention periods, and store data and train models close together. If data must move, compress it.
About this talk
As more organizations align their business with sustainable practices, it is crucial to understand the impact of your IT resources. In this session, you will learn how you can use the cloud in your transformation to more sustainability. We'll take a look at the shared responsibility model for environmental sustainability and share best practices for architecting for sustainability. This session dives deep on architecture patterns that can help you make your infrastructure resource-efficient, and therefore more energy efficient, with a focus on user patterns, data and hardware patterns, as well as software design.
Why sustainability is harder than digital transformation (60 s)
Sustainability of the cloud vs in the cloud (71 s)
Greener data habits for machine learning (71 s)
Speaker 1: All right, it's half past. So let's start this session and welcome everyone to my session on architecting for sustainability in the cloud. My name is Katja and I am a solutions architect with Amazon Web Services focusing on the topic of sustainability. And within that, I am a solutions architect with Amazon Web Services, focusing on the topic of sustainability and within that, I am a solutions architect with Amazon Web Services, focusing on the topic of sustainability.
Speaker 1: And within that I support my customers within their whole sustainability transformation. So that means, for example, in terms of reporting, but also because I'm a solutions architect, how can I actually optimize my workloads and my architectures for sustainability and what kind of best practices exist here. Speaker 1: Oh, sorry, I just have a small issue with my browser here. Okay. But I'm back. All right. And yeah, if you have any questions, feel free to put that in the chat and I'll be checking that out.
Speaker 1: Okay, let's jump right into it. So before we'll dive into architectural best practices, I wanted to start with a brief overview of the term sustainability. What do we actually mean with that? Speaker 1: Then talk a bit about sustainability at AWS, just to give you some kind of overview, what kind of initiatives we have here. And then, of course, talk about the different best practices on how can we optimize workloads for sustainability.
Speaker 1: So I'm sure all of you are aware of this term sustainability and how it is actually important for us. Speaker 1: Okay, let's look okay here. Yeah. But I anyways brought the definition of the United Nations with me here just so we're all on the same page. Speaker 1: So the United Nations define sustainability or sustainable development as meeting the needs of the present without compromising the ability of future generations to meet their own needs. So I think this is a very, Speaker 1: It's an abstract definition, but it comes to the point, of course, on what sustainability actually is about sustainability is a broad term. So it means not only decarbonizing our business operations, but it's also around water conservation, responsible employment or circular economy. So really a very broad term. And today we'll focus mainly on environmental sustainability and initiatives that address.
Speaker 1: That address decarbonization and how we as technologists can bring our part to optimize our applications for sustainability. Speaker 1: And sustainability is of course, yeah, important to all of us. And in this, in this technical field, we probably also see a shift towards a sustainability transformation. I think that's always quite interesting because it's technical. Speaker 1: We're so familiar with the term of digital transformation. We've done that in the last decade. We've been able to connect with our customers with everything that we make, and we've been able to transform our businesses. But now we really, we're embarking on the sustainability.
Speaker 1: And then we're using the same tools as a digital transformation. And we're using the same tools as digital transformation, but focusing this time on sustainability. Speaker 1: And I think that's always quite interesting because as technologists, we're so familiar with the term of digital transformation. We've done that in the last decade. We've been able to connect with our customers with everything that we make. And we've been able to transform our businesses.
Speaker 1: But now we really, we're embarking on the sustainability transformation using the same tools as a digital transformation, but focusing this time on sustainability. Speaker 1: And even though we see this shift, we should still see these two transformations together because actually companies that have understood both digital and sustainability transformation and are driving innovation. Speaker 1: And exactly the intersection of digitalization and sustainability are much more successful in the market. And here you may have heard the term of the twin transformers or twin transformation.
Speaker 1: But there definitely is a very important difference between sustainability and digital transformation. Speaker 1: And you may have already thought about that because sustainability, of course, takes decades. Well, digital transformation. Speaker 1: It just happens in maybe months or years. So the really important challenge here is with sustainability is how can we achieve an outcome that is decades away while we actually know in order to get there, we need to accelerate the process today.
Speaker 1: So this is really such a big challenge in itself. And here I just brought some other challenges. Speaker 1: Some examples that our customers face, but also we as AWS are facing. Speaker 1: That is how can I actually identify carbon emission hotspots, but also how can I measure and reduce the environmental footprint, including energy and waste water.
Speaker 1: And how can I collaborate with others in my value chain to actually reduce emissions. And I, as a solutions architect focusing on sustainability, support my customers. Speaker 1: With regards to these different challenges. Speaker 1: And now when we look at specifically the cloud and now I will talk about AWS, of course, because that's where I work. But in general, these, these concepts can be applied to different, different cloud infrastructures or cloud providers.
Speaker 1: So here are the three general ways that we can collaborate with sustainability transformation in mind. Speaker 1: And that is, first of all, migrating workloads to the cloud in order to take advantage of higher energy efficiencies and high utilizations of cloud infrastructure. Speaker 1: Then secondly, optimizing workloads for a further conservation of energy. And this is what we'll be talking about today.
Speaker 1: And then lastly, transforming other areas of your business with the use of technology and data to reduce carbon emissions across all of your operations and not just IT. And oops, that was a bit too fast. Yeah. Just because of the limited time we have today, we'll focus on the first two parts. Speaker 1: So let's first of all and talk about migrating. So migrating workloads to the cloud can lower your carbon footprint compared to the typical on prem data center. And I think that is just important to, to understand. I was at a conference a couple of weeks ago and someone asked me, we are already, or we already did all these optimizations for resource efficiency 40 years ago.
Speaker 1: But what has actually changed now. And of course, with the cloud and with this more centralization of a data center that multiple companies are using and where the cloud providers are optimizing for energy efficiency. We can, of course, reduce the carbon footprint just by itself. Instead of having a lot of small little data centers. Speaker 2: That are not as energy efficient and that are also not as highly utilized. But I would like to just give you some more insights onto the different emissions that we actually have in a data center and then talk about how we are reducing these.
Speaker 1: You may be familiar with the greenhouse gas protocol and the different scope one, two and three emissions, but I'll anyways explain these. So no worries. If you have an idea. Speaker 1: So no worries if you haven't heard of these yet. So what kind of emissions do we have in a typical data center? We probably have a backup generator in case of power outages or to stabilize the grid. And these direct emissions contribute to our scope one emissions. So the direct emissions.
Speaker 1: We're also consuming an electricity mix coming from the local grid as well as from our investments into renewable energy projects. So these carbon emissions contribute to our scope two emissions. As you can imagine, these take up really a significant portion of our overall greenhouse gas emissions. Speaker 1: And scope three is then really everything else. So the data center itself, which needed to be built from concrete and steel, the equipment that we have inside the data center. So electronical equipment racks and servers.
Speaker 1: All of these upstream emissions. So the production of these equipment and data center itself or the transportation of these different equipments, but also the downstream emissions. So things like how do we retire our servers when they come to their end of life in terms of recycling or refurbishing. Speaker 1: So as you see, all of these details have different carbon emissions attached to it. And all of this needs to get to net zero carbon by 2040, which is our long-term target at AWS.
Speaker 1: And here, just some initiatives. We started to move closer to this goal. So first of all, of course, scope two emissions. So our, the energy that we're consuming here, we are investing heavily in different renewable energy projects. Speaker 1: And Amazon is actually the largest corporate buyer of renewable energy in the world. And we are on a path to powering all of our operations with a hundred percent renewable energy by 2025. And that means not only AWS data centers, but also corporate offices or fulfillment centers.
Speaker 1: With regards to scope three, we can think about new, um, new initiatives, new innovations. And for example, Speaker 1: The construction of our data centers, like the use of low carbon concrete. And there are also several, um, Speaker 1: Goals on water usage effectiveness or power usage effectiveness. Speaker 1: And another really interesting part, um, I think is the chip development. So here we partner with arm to develop a new family of instances that are more energy efficient.
Speaker 1: Just last year we have brought out the Graviton three powered instances, which are, um, more energy efficient. They are 60% more energy efficient than comparable instances. Speaker 1: So just by picking actually this instance type, you can already improve the, the energy efficiency and the carbon footprint of your applications so much. Speaker 1: So with all of these initiatives, we are reducing our carbon footprint, um, in order to give you access to a cloud that is most sustainable.
Speaker 1: And here just to quantify this a bit. So in a benchmark of typical on-prem data centers, co-locations and cloud providers, they found out that for example, in Europe, our AWS infrastructure is up to five times more energy efficient than the typical on-prem data center. Speaker 1: Which also means that just by migrating a workload to the cloud, a customer can reduce their carbon footprint by nearly 80%.
Speaker 1: And here we didn't even yet talk about these architectural optimizations that you can, can take into account. Speaker 1: And where does it come from? So that comes from the higher energy efficiency of both, um, servers and data center facilities, and also the higher utilization of, of our servers. Speaker 1: All right. And then here, if you're familiar with AWS, you may have seen the shared responsibility model for cigarette for security. And this also, um, is applicable for sustainability.
Speaker 1: So sustainability is a shared responsibility between assets, AWS and our customers. So we, as AWS are of course responsible for the sustainability of the cloud. So our data center. Speaker 1: So our data centers, the global infrastructure, the use of renewable energy or partnerships with processor producers. Speaker 1: On the other hand, our customers are responsible for the sustainability in the cloud. So your workloads essentially designing your applications for sustainability and resource efficiency.
Speaker 1: Optimizing data usage patterns or software design. And we'll now jump into the best practices on how you can actually optimize your workloads. Speaker 1: So first of all, um, let's talk a bit about what that actually means. And based on what metrics I can optimize my, my workloads for sustainability. Speaker 1: When we talk about optimizing for sustainability, we focus on the energy reduction and the optimization of efficiency across all components of our workload.
Speaker 1: Essentially using the least amounts of resources needed and using these to the fullest. Speaker 1: And now you may thinking that my application teams are already optimizing their applications for different metrics, such as cost or response time. So what kind of metric can I use for sustainability? Speaker 1: And here I want to quickly talk a bit about proxy metrics before going into the recommendations.
Speaker 1: Proxy metrics are metrics that serve as fine granular substitute for CO2 emissions, because of course we can look at the carbon footprint and we should look at the total carbon footprint that our, for example, AWS usage has. Speaker 1: And you can do this with the customer carbon footprint tool, but, um, this number of total carbon emissions is not really tangible for an optimization because we just want to reduce it, but we don't really know how.
Speaker 1: So we should also look at other metrics that actually make this a bit more tangible. Speaker 1: And here you already heard that optimizing for sustainability is in large parts about resource efficiency. Speaker 1: So first of all, we can look at the resources that we are using in terms of compute storage and data transfer.
Speaker 1: In general, we want to reduce the cost as in the cloud. We have a pay per use model. Speaker 1: In order to now normalize the resource usage to a unit of work, we should also look at the resources per unit of work and reduce this number. Speaker 1: So that could be, for example, how many resources do we need to serve 10,000 customers?
Speaker 1: Or how many vCPU hours do I need to handle a hundred orders? Speaker 1: So essentially putting our resource usage in proportion in order to compare this KPI to previous month and fluctuating usage. Speaker 1: We can also look at increasing the utilization per resource. Speaker 1: And now if we talk about networking here, the utilization in terms of sustainability is a bit hard to measure.
Speaker 1: So we can optimize the overall network utilization by metrics like data size or distance my data has to travel. Speaker 1: And here it's just important to keep in mind that there may be different and trade offs. Speaker 1: With common non-functional requirements, such as data retention response time or availability. Speaker 1: So if you think of storage, just think about do I need access to all of my data in milliseconds?
Speaker 1: Okay, so now we have talked about the metrics. Speaker 1: So let's continue on to the best practices. Speaker 1: So how can I actually optimize my workloads? Speaker 1: And here again, you may be familiar with a well architected framework, which just provides different best practices to evaluate and implement architectures on these different pillars.
Speaker 1: And last year we have announced the new sustainability pillar, which also includes best practices to design and implement and operate workloads in a sustainable and resource efficient way to essentially further reduce your energy usage. Speaker 1: And this is now such an important step to seeing sustainability as the same non-functional requirement as cost or security.
Speaker 1: When you're already designing an application. Speaker 1: The best practices comprise these five different areas. Speaker 1: First of all, user behavior patterns. Speaker 1: So looking at how are your users actually using your workload? Speaker 1: Where are they located? Speaker 1: When do they use it in order to drive decisions on your architecture?
Speaker 1: And software and architecture patterns. Speaker 1: We may look at how energy efficient is the programming language that you're using because there are a lot of differences here already. Speaker 1: But also how have you architected your code? Speaker 1: Which areas of your code consume the most resources? Speaker 1: Hardware patterns is essentially about have you picked the right hardware for what you're trying to achieve?
Speaker 1: So here again, talking about Graviton, but also talking about right sizes. Speaker 1: And next to hardware patterns, we have data patterns where also often see a huge area for optimization. Speaker 1: So gaining an understanding about your data, classifying it and using the right storage technology. Speaker 1: And then lastly, in development and deployment processes, looking at how are we developing our applications?
Speaker 1: And how are we utilizing our built environments, for example? Speaker 1: And just to make this a bit more tangible, I brought an example. Speaker 1: So let's just imagine that we have a machine learning workload that we want to optimize. Speaker 1: As you may know, or can imagine, training artificial intelligence and machine learning workloads, of course, uses a lot of energy.
Speaker 1: But we can also use machine learning to optimize for resource efficiency or fight the effects of climate change. Speaker 1: So let's just imagine we are building a machine learning model to forecast extreme weather events. Speaker 1: And here on the right, you can see the machine learning lifecycle, an iterative process with the six phases of developing a machine learning workload.
Speaker 1: So that starts at the identification of a business goal, framing this as a machine learning problem, collecting and preparing data, developing and training a model, and then deploying and monitoring it. Speaker 1: And throughout all parts of this life cycle, we can, of course, apply the best practices of the sustainability pillar. Speaker 1: So let's just exemplary look at the data patterns.
Speaker 1: So here, oops, sorry. Speaker 1: Here already at the data processing stage, we can evaluate if we can avoid data duplication by using existing publicly available data sets. Speaker 1: So here, for example, the Amazon sustainability data initiative provides access to different data sets for free, such as weather data or satellite data. Speaker 1: We should also classify our data and implement data lifecycle policies in order to move data to energy efficient storage and use appropriate storage tiers, essentially moving data to colder storage tiers the less frequently we use it.
Speaker 1: We should also define data retention periods and delete unused data. Speaker 1: And lastly, look at the data movement across networks. Speaker 1: So here we should, of course, minimize the movement over the network by storing our data as close as possible to our producers and training our models as close as possible to our data.
Speaker 1: If we still need to move data over the network, we can apply compression in order to reduce the size of the data. Speaker 1: So these were just some examples in order to make it a bit more tangible. Speaker 1: If you're interested to learn more about this, then please check out the well architected pillar and white paper.
Speaker 1: And yeah, I know I've gone a few minutes over. Speaker 1: So here, just as a summary, what I would like you to take away from this session. Speaker 1: So first of all, of course, set a sustainability goal and introduce mechanisms to assess these proxy metrics per team and per application. Speaker 1: Because of course, what we can't optimize what we can't measure.
Speaker 1: Then look at these different best practices and identify areas of improvement. Speaker 1: And then prioritize findings, make changes and assess the impact again by looking at the proxy metrics. Speaker 1: All right. Speaker 1: Yeah. Speaker 1: Thank you so much for listening in. Speaker 1: If you would like to know more about this topic, you can definitely reach out on LinkedIn.
Speaker 1: And yeah, I hope you have an amazing session, amazing day and lots of interesting sessions. Speaker 1: Bye everyone. Speaker 1: Thank you.
Transcribed automatically; names checked against the speaker record. Select a time to jump there.
Article based on this talk
Who is responsible for making cloud workloads sustainable?
Katja Philipp of AWS explains why efficient data centers are only half the job and how architects can measure and cut the resource use of their own workloads.
In short
Sustainability in the cloud is a shared job. Katja Philipp, a Solutions Architect at Amazon Web Services, argues that while the provider is responsible for efficient, renewably powered data centers, customers are responsible for how their workloads use those resources. She shows how proxy metrics and the five best practice areas of the AWS Well-Architected sustainability pillar turn that responsibility into concrete design decisions.
Key takeaways
The shared responsibility model for sustainability
Best practices for architecting for sustainability
What sustainability means in the IT and cloud context
Sustainability is slow, so the work has to start now
Philipp opens with the United Nations definition of sustainable development and notes that sustainability is a broad term, covering decarbonization, water conservation, responsible employment and the circular economy. Her focus is environmental sustainability and what technologists can do to optimize applications.
She frames the current shift as a sustainability transformation that uses the same tools as the digital transformation of the past decade. She suggests the two belong together, since companies that drive innovation at the intersection of digitalization and sustainability are, in her words, much more successful in the market, an idea sometimes called twin transformation. The difference is time. "So the really important challenge here is with sustainability is how can we achieve an outcome that is decades away while we actually know in order to get there, we need to accelerate the process today." (4:23)
Where data center emissions come from
Philipp outlines three ways the cloud supports a sustainability transformation: migrating workloads to benefit from more efficient, highly utilized infrastructure; optimizing workloads to save further energy; and using technology and data to cut emissions across the wider business. The talk covers the first two.
To explain migration, she walks through the Greenhouse Gas Protocol scopes as they apply to a data center. Backup generators create direct scope one emissions. Electricity from the grid and from renewable energy investments falls under scope two, a significant share of the total. Scope three covers everything else, from the concrete and steel of the building and the servers inside it to transport and how servers are recycled or refurbished at end of life.
As the speaker cited, AWS targets net zero carbon by 2040, Amazon is the largest corporate buyer of renewable energy in the world, and it is on a path to power its operations with 100% renewable energy by 2025. She also points to low carbon concrete and Graviton3 instances, which she says are 60% more energy efficient than comparable instances. She cites a benchmark finding that AWS infrastructure in Europe is up to five times more energy efficient than a typical on-premises data center, meaning migration alone can reduce a carbon footprint by nearly 80%.
The shared responsibility model for sustainability
Familiar from security, the shared responsibility model applies to sustainability too. AWS is responsible for sustainability of the cloud: data centers, global infrastructure, renewable energy and processor partnerships. "On the other hand, our customers are responsible for the sustainability in the cloud." (12:30) That means designing applications for resource efficiency, optimizing data usage patterns and choosing good software design.
Optimizing for sustainability, she explains, is about reducing energy use and improving efficiency across every component of a workload: "Essentially using the least amounts of resources needed and using these to the fullest." (13:43)
Proxy metrics make carbon reduction tangible
Teams already optimize for cost or response time, so what should they optimize for here? Philipp recommends tracking the total footprint with the Customer Carbon Footprint Tool, but notes that a single emissions number does not tell a team how to reduce it. Proxy metrics act as fine-grained substitutes: resources used across compute, storage and data transfer; resources per unit of work, such as vCPU hours needed to handle a hundred orders; and utilization per resource. For networking, data size and the distance data travels serve as proxies.
She warns of trade-offs with non-functional requirements such as data retention, response time and availability. "So if you think of storage, just think about do I need access to all of my data in milliseconds?" (16:06)
Five areas of the sustainability pillar, applied to machine learning
The Well-Architected Framework gained a sustainability pillar, which Philipp sees as a turning point. "And this is now such an important step to seeing sustainability as the same non-functional requirement as cost or security." (17:00) Its best practices cover five areas: user behavior patterns (where and when people use a workload), software and architecture patterns (including the energy efficiency of the programming language), hardware patterns (right sizing and choosing hardware such as Graviton), data patterns, and development and deployment processes.
To make this concrete, she imagines a machine learning model that forecasts extreme weather events and applies data patterns across the machine learning lifecycle. "Here already at the data processing stage, we can evaluate if we can avoid data duplication by using existing publicly available data sets." (20:05) She points to the Amazon Sustainability Data Initiative for free weather and satellite data, then recommends classifying data, using lifecycle policies to move it to colder storage tiers, setting retention periods, storing and training close to the data, and compressing data that must travel.
Her closing advice is a loop: set a sustainability goal, assess proxy metrics per team and per application, identify improvements against the best practices, prioritize changes and measure the impact again.
Watch the talk
Architecting for Sustainability in the cloud, Katja Philipp, 23 minutes: https://www.womentech.net/video/architecting-sustainability-in-cloud-katja-philipp
About the speaker
Katja Philipp is a Solutions Architect at Amazon Web Services based in Munich, Germany, who joined AWS in September 2020 through the TechU Graduate program after a Master of Science in Information Systems. She supports customers in the Power & Utilities vertical with best practices for their cloud journey and is passionate about how technology can help solve current challenges for a better future.
Questions this talk answers
What is the shared responsibility model for sustainability in the cloud?
How can teams measure sustainability when a total carbon figure is hard to act on?
Why can migrating to the cloud lower a carbon footprint?
Which areas does the Well-Architected sustainability pillar cover?
How do data best practices apply to a machine learning workload?
Drafted with AI from the talk transcript. Editor review pending.
AWS Well-Architected sustainability pillar white paper
AWS Customer Carbon Footprint Tool
Amazon Sustainability Data Initiative
Cite this talk
Katja Philipp. "Architecting for Sustainability in the cloud." WomenTech Network, 2022. https://www.womentech.net/video/architecting-sustainability-in-cloud-katja-philipp