Running sustainable real-time advertising workloads in the cloud

hi, i'm stephanie lazer. i'm on the advertising and marketing technology team at aws. and today we have a very special presentation for you. we're gonna talk about sustainability at aws.

we're committed to sustainability of the cloud and helping our customers with sustainability in the cloud. uh today we're gonna have anal elect and amazon ads. uh give a presentation.

first up is rakshit upal who's gonna talk about uh real time advertising, workloads in the cloud.

thank you, stephanie for the intro. i'd like to start with a little story when i was a little kid. my mom used to always remind me to turn off the lights as i left the room or to close the back door as i go out to the yard was super annoying at the time. but what she was really asking me to do was to be less wasteful. and here i am today to talk to you about how we can reduce waste and be more sustainable.

hi, my name is rakshit. i am a senior manager at amazon ads. i sure hope you're having a good time at re invent amongst other things. my team owns the sustainability initiative. at amazon ads simply put rebuild and drive mechanisms to help tech teams and ads recognize and reduce our impact on the environment.

so let's see what we've got in store for you today. we want to start with a short intro to amazon ads. then we'll cover the shared responsibility model that was just alluded to with aws. we will then dive right into what amazon ads is doing. starting with our industry partnerships. we follow this. let me keep moving through. we will follow this up by our internal strategy which is three simple steps, measure educate and reduce. we will discuss more about this as we go through the talk to wrap things up. we're going to share how ads has innovated in this space and things that we've experimented with, followed by learnings and some takeaways that you can use to make a real impact.

so let's get started with the quick overview for amazon ads. amazon ads reach customers and audiences across amazon and third party properties. there's alexa freebie music, you name it. this next slide provides a view of the various products. amazon ads has to offer each of these products is geared towards our mission which is to help brands design a experiences that delight customers and deliver meaningful results with their insights and reach. they help connect advertisers with the right audiences in the right places, both on and off. amazon.

next, we're going to talk about the advertising life cycle we're going to be coming back and referring to this a few times. so let's go slow to the left. you see two personas, there's advertisers and there's three p developers or folks who work with a ps, they integrate with our campaign management systems and a ps to build and manage their campaigns. all of these campaigns then get served via our hyper scale a server and bidding systems. and from there while a serving is happening, we record any clicks impressions or views and we aggregate budget spends in a real time manner from there. we then store all of this data in massive data lakes and then use ml techniques to build sophisticated models to improve a selection and relevance. final piece is that we return measurement data back to advertisers through our campaign management systems completing that life cycle.

with that. let me give you an idea of the scale that we operated. some of our a systems, manage millions of tps, process a requests in low latency situations, operate hundreds of highly available services and this scale results in tens of billions of impressions worldwide and several petabytes of storage.

today we mature insights that come from operating at this scale so that you can benefit from it before we move any further. let's first level set on terminology. what is sustainability?

sustainability is a discipline that addresses the long term impact of our activities on the environment. sustainable development is one that meets the needs of the present without compromising the needs of the future and the practice of sustainability. and this is what we adopt is to understand the impact of the energy we consume, measure this across the life cycle and apply best practices to reduce the impact of their use.

so let's go deeper first up as was alluded to earlier by stephanie, there's the shared responsibility model and that's kind of what surrounds our sustainability sort of framework. there is aws that is responsible for sustainability of the cloud and then we can help by using the minimum amount of hardware we need to meet our needs in the cloud. we believe that by working together, we can really scale our impact.

let's talk about what amazon ads is doing. we are investing internally as well as externally outside of amazon externally, we recently announced a partnership with ad net zero, which is an industry coalition that accounts for more than 40% of world's ad spent through its members added zero is now a signatory to the climate pledge, which is a fantastic step forward internally at ads. and this is what i was alluding to when we're going through the agenda. we follow a simple three step plan. we had to figure out a way to measure our footprint, then educate our teams and finally find ways to reduce that footprint.

so let's talk about measuring, we identified that our total footprint ultimately consists of two things. there's things like building, there's travel, there's scope three, there's emissions that come from serving ads on three p websites or third party websites. and the second bucket that we control more directly is our tech footprint on aws. today, we are going to focus on this second bucket.

so how did we actually measure our footprint? we did this in correlation to the life cycle we spoke about earlier. we started by identifying all aws accounts used by amazon ads. any guesses out there, how many accounts we might have had? everything's 500 perhaps, maybe 1000. any guesses at all, we had north of 80,000 accounts for each aws account. we were able to get our carbon footprint and in the same way that aws makes it available for all aws accounts, we then aggregated this data at fleet level or workload level and there you have it. we got cf per workload or per fleet. now these workloads in our case map back to the life cycle we discussed earlier. so it was a fairly easy way to aggregate our footprint.

what did we find out of that? we found that four products very unsurprisingly accounted for 99% of our storage of our carbon footprint. there's the compute side of things, which is amazon ac two and other variants of two and then the storage around three ebs dynamo, etcetera.

so now that we were getting a better handle on our footprint and where we needed to focus. we moved to the next step and the next step was to educate at amazon when we want to get an engineer thinking, what do we do? we run a hackathon and that's exactly what we did earlier in the year, we got 50 plus teams participating there. and they were fantastic ideas.

here are two that i would like to share with you today at our scale where we can have hundreds if not thousands of hosts in any fleet. we have to find an easy way to perform upgrades and also that needs to be fully automated. and that's what one of our teams did. so we could then upgrade our fleets iteratively to more newer versions of graviton or other more efficient instances in an automated and safe manner.

the next area was around, right, sizing hardware, we had a number of ideas there, but i want to talk to you about one of those, one of our ml teams innovated in that space and showed us that there are some less complex models that do not need to run on the more carbon intensive gp u instances and perform equally well on cpu based instances. they ran a po to show this and we'll be figuring this out and implementing this down the track.

so now that we are learning all these things, how do we make these insights useful? this is the part where we started to focus on reduction. they started with low hanging fruits such as avoiding waste, moving to more carbon efficient instances, followed by workload optimization. we're going to talk to all three aspects in the next few slides.

like i told you the story earlier about my mom

This is right here is the same practice applied to our EC2. We started with identifying unused EC2 instances across all FADs. We built a tool that retrieves unused hardware data across all of those 80,000 accounts I mentioned earlier and surfaces on demand capacity reservations or ODCRs that are unused in any given fleet.

Now, there are some very valid use cases too where you will have some unused hardware such as when you're preparing to scale and we do that for Prime Day and other events like that so perfectly reasonable. But when it starts to go beyond, let's say 14 days or so, we start to get buried and we can take those teams, talk to them, figure out what's the best way to optimize since the launch of this through this dedicated effort that we've driven.

There's been meaningful change that has led to us being more frugal. We have driven a 60% reduction on an average in unused hardware in the last six months from the life cycle. We looked at earlier.

This ad serving and measurement are two of our largest and most complex set of systems. One such large ad serving system capable of handling millions of TPS at low latencies pushed the limits of optimization by migrating the entire EC2 fleet to Graviton in all regions. This gave them a cost benefit of about 35% and reduce carbon emissions by 44%. They went from, I believe, C5 6Gs, they went to C6Gs from C5s. This reduction has created more headroom for greater innovation for them.

There's another large measurement system that moved about 35% of their workloads to Graviton and that cut their emissions down to in half. We had a very large ads of cashing a serving cash fleet that ran a PoC to leverage CSTD compression. Our test results indicated that CSTD compresses our cash payloads by about 20% on an average. Now that's what we experienced. We know that this is something that is useful and results and you know, numbers may vary as you're on your own experiments. But what this did was it helped, it's going to help us reduce the number of hosts that we need in that cash lead.

There was one thing related to two that you should take away from this session. It is this slide. This is ee essentially the derivative of what we found over our journey. Now, these like I said earlier are our general observations and results are going to vary based on fleet sizes, workload, CPU utilization and a whole bunch of other factors. But that gives you the general idea as an example, going to R6 6Gs is going to save you 46% roughly over R5 and 16% over R6.

Next, let's talk a little bit about storage in the age of storage. We would like to share two insights use S3 intelligent tearing. One of our ad event stores recently moved to three intelligent tearing and they were, they managed to get both cost and carbon reduction. We also built an unused tool and an automated tool that can detect unused data sets that are no longer being queried and it cleans them up along the same thing of reducing waste.

Here's a quick summary of the, of the things that we've done in that space. There's law compression, we spoke about GSTD, there's life cycle policies, you know, use intelligent tearing, you got to set retention policies in S3. Then along the lines of reducing waste, you know, there's deleting unused and unattached EBS volumes and finally, one area that we didn't quite go in too much depth earlier on is around using common formats like Parquet, which are more storage efficient than other formats.

I would like to finally talk about one simple practice that we'd like to encourage within Amazon Ads, watch your attachments in emails, use some sort of network storage perhaps like WeTalks, imagine if you send one email with a meg attachment out to 100 people and they proliferate that over to another 100 people just think about the data replication that is resulting from that by simply storing it in a, in a, in a in one network location, you're going to be saving a lot of space.

There's another very simple method which is, which might sound fairly straightforward, but it helps turn off your laptops when they're not in use, extends your battery life and also reduces your electricity consumption. These simple things done at scale can have a really large impact.

So before I move to the next part, before we move to the next part of the presentation, um a quick summary, focus on reducing waste, try to adopt more carbon efficient hardware and focus on workload optimization. There are plenty of other things you can do this. You know, if you're a Java based shop JDK 17, you know, and, and a few other things like that will also help.

So with that, thank you all for listening. I would now like to call upon stage, John Driscoll from Analy. He's going to share their journey. Thank you, John. Thank you.

Hi, everybody. I'm uh John Briscoe with uh I'm the director of data and operations at intellect. Uh I am going to warn you that I'm a fast talker and uh I've got 20 minutes and lots of content to go through. So uh I'll, I'll be available afterwards in the hallway uh or if we have questions to answer at the time. Uh I can't speak for Michael, but maybe he'll be there too. Uh he's probably more knowledgeable.

So who is intellect? So over a decade ago, intellect was created by omnicom uh to reimagine the future of marketing. Now, we're a 4000 strong global specialty services company delivering data driven capabilities, proven to drive better marketing outcomes for our clients.

Um omni which is the flagship operating system used by marketers to orchestrate better outcomes is supported by our data platforms and it requires more than eight trillion impressions, refreshed daily, over 10 million ad units furnishing capabilities for over 41 markets. All this comes together to drive our need for 51 petabytes of data online at any given time.

So I'm just reading through that all i i read through that uh just so everyone could get an idea of the scale of data we're talking about here.

So um this particular slide i kept burying uh because it makes a great like m night shyamalan style twist at the end of hey, we're the most of this talk is gonna be about how to save money. Uh but it also ties in very well with how to save carbon emissions. And so this slide i kept burying cause it was like, aha look, you can save emissions, but really this is the heart and soul of it.

So if you take anything away from it, it's this slide. So um i've tried to put this in order of uh uh ability to transform as well as difficulty to implement. Um and, and all of these, all of these things listed here will come with not only a major carbon emission impact uh for your organization, but also uh with a, a cost savings which makes it really easy to um sell to both the org and to your finance team.

So this sets the stage, right? So, um w when you come into this, it's very important, i think in the cloud, especially when you're talking about complex uh uh implementations such as things that have to deal with carbon emissions that you are bringing in both the org side and the finance side along with tech um to, to, to walk through what you're trying to do because especially when you start talking about a scale, things get really complex and your ability to uh blow up the system, which i'll get to in a little bit later uh can be uh very problematic.

So having a way to bring the finance team in the organ to say, hey guys, this is why this happened. We knew this was a thing, you know, here's what we're doing to solve it, et cetera. It is very important. Uh for these slides, i'm gonna have these uh faces representing uh each of the orgs i tried to get don draper. Uh because when you think of ad tech, you think of me and don draper. Uh, but the, uh, unfortunately you get me instead.

So, um, so like many organizations, um, that are starting their cloud journey off, we started off, uh in the cloud and it's very easy to sell at the time. Uh, when you're starting off to the finance team because they're like, what's our start up cost? What's the SAN gonna cost us? What's a server? And you're like, nothing. It doesn't cost anything. And they're like, yeah, they're excited then you start doing stuff and it immediately they're upset because you're, you're costing money.

So uh uh i'm mostly joking about this. It's um um a as long as you're bringing the finance team in with that at the start, they really totally understand uh what's going on and they're, they're, they're fairly happy to, to get how growth is working.

So, um starting off in 2009, we had just the basics up through 2014. We were doing uh just S3 buckets, EC2s uh RDS, basic RDS. It wasn't even Aurora at the time. Um and Amazon Redshift, Amazon Redshift is really transformative for our org. Uh and over the next years, this allowed us to evolve a holistic uh platform for data scientists. And in 2018, start to empower end users with holistic marketing platform.

Uh and all of this comes together to, to surface that we had really the basics. Uh and we grew um exponentially over time, just with data and capabilities and to support this kind of thing, it takes an enormous amount of data. Uh and so what that led to was this mm charts, everybody's excited about charts.

So what that led to was um these basic services were, were, were growing at a scale that was hard to justify to our finance team, right? So we're saying, you know, the the 15 petabytes of data online at that scale S3 is problematic. And so um we, we were really struggling, you can see during the in the bulk of this to both control costs, which also means controlling emissions.

"Um and, and also be able to provide the data we needed to support the platform. Meanwhile, uh our, our parent company, Omnicom starts to make um uh uh uh pledges to reduce energy consumption and all of our clients are becoming more carbon conscious. So it would be nuts for us not to be focusing also on, on carbon emissions on top of the fact that we live on the same planet that everybody else does. And so we actually do care, right? Like there's a, there's a reason to care. And so that brings a new player into the, into the mix, which is earth. Uh and so, so now you're, you're you're trying to justify to both the org and the finance team what you're doing from a cost perspective, but also uh from a carbon emissions perspective.

Um and so this is what our original stack looked like. Uh and in the early days of this, we were talking with Amazon about how do we control for emissions, how do we, how do we figure out what we're doing? And um the, the response from Amazon was essentially focused on cost reductions. Uh like Rex said, it's, it's uh turning the light off, you know, turning things off that reduces your cost and that ties directly to carbon emissions. And so that, that made a lot of sense to us. And so, uh and also uh it was really easy to sell the finance team as we're like, hey, we wanna reduce emissions. It's also gonna save us a lot of money and they're like, yeah, ok, you got it, you sold us.

So um so that was what we were doing. This is what the stack looked like. This has continued to grow. Uh obviously, we didn't stop in 2020 2020 on to 2023. Uh we, we rolled out Omni customization, which just required an enormous amount, more data and then into 2023 anybody who's dealing with tech right now knows that there's uh a significant generative AI uh explosion which of course takes GPUs and more compute power and more availability, which is more data and more cost and more emissions.

Um so this is just continued onward. As, as I think it does for probably most organizations. So the first thing we did was try to tackle this from uh the point of view of, of our, of our ad tech. So from the ad tech side of things, uh we joined with organizations such as Climate Partner, Scope 3 and, and others to create uh things like our carbon emissions scenario planner, uh which we integrate into Omni our main, our main platform, um a scenario planner that it's a scenario planner that enables planners to optimize plans for lower emissions. While also considering clients budgets and performances, it allows teams to compare a previous planned media plan with a sustainable uh optimized media plan and recommend uh a media mix shift in favor of sustainability.

Um so that's how we target it from or that's how we tried to reduce emissions um and control for costs from an ad tech point of view. But I work on the tech team. And so like this isn't uh aside from implementing this stuff, this isn't something I could very well control. So from a tech perspective, we did this which is lots of arrows. So uh we moved uh to on demand wherever possible and uh reserved wherever predictable, right? And so when I say on demand, in this case, I'm saying, you know, from our original stack, you will remember um we had an enormous amount of EC2 was basically our only compute. And so what we did is we for EC2, we moved that off to, to things like Lambda, Fargate Batch, Step Functions, etc. Also on demand during this time period um and unreleased uh AWS uh or, or Redshift Spectrum, which is the query layer that exists on on Redshift. And so that was really transformative to us as well because we were able to reduce a lot of the size of our Redshift clusters which reduces cost by a lot and shift all of that cost over to on-demand query uh patterns which uh it really did change the, the, the, the face of the tech team.

So um beyond on the other, on demand is like um moving uh our, our processes to EMR wherever possible. Uh and um uh Aurora on demand as that came available, there's now also Redshift on demand, I think that was released two years ago. Uh and, and is we're working with that too and seeing how that works and we've had some success there as well. So this was what our original stack looked like. And this is that chart of cost uh with the bloated middle part. Um and once we implemented these, we got this which is a clear trend line of cost reduction uh uh which you'll remember ties to emissions uh starting basically uh in 2020. Uh and I also need to call it during this time that our, our, our growth was at, at the same pace. It had always been like nothing slowed down. Um we were growing in both uh usage patterns and data consumption, right? Um and so the, the, the, the fact that we were able to consider this in this way was really cool, I thought and um I left this in here. So you guys knew this was real data because it looks better if there's not that spike. Uh but there is a spending anomaly there.

So, one of the problems with moving to on demand is you have to be a little bit concerned about uh usage patterns uh by data scientists or whoever that might be making use of something, not a nefarious way. Uh but you know, having a query that, that doesn't run optimal Spectrum or whatever uh that can cause that spike. So that, that spike is a real spike. Uh and I'm sure I had to answer for that uh when it came through.

Um but then during this time, then Amazon released uh the AWS carbon emission calculator, which was incredible. So this gave us a real window outside of billing to say, how much is this costing us uh in emissions. So we started way back when of just saying, ok, cost equals emissions. That's great. Then this came out and it really transformed the way that we look at all this. So we could stop looking at just cost and we could say demonstrably like this is actually emission data. This is what we're seeing uh when we make this shift from on demand or when we reserve.

Um so that gave us charts like this chart um which uh was proving basically that AWS was right control for cost also controls for emissions. So this chart uh the orange line is our EC2 emissions um in metric tons of carbon emissions. And uh the other line is basically just Redshift um in in Redshift that you'll remember we reduce the size of our cluster, which is that big spike down by, by leveraging Spectrum. Um but we also predominantly could only make use of like reservations. Um so it doesn't, you don't get the same kind of emissions and, and cost savings as you would as moving to uh on demand, things like Lambda and Batch, etc. Uh but we did see a pretty big uh uh dip. Uh and again, this was with more data and more usage. So incredible.

Um so then we wanted more, right? So now the org is like, hey, we can see this, this is great. And so then we started talking about things that could save us, still save us money but less money uh and have an impact on our our carbon emissions. So we started talking about making movements uh over our EC2 fleet over to the Graviton processors. Uh the Graviton processor is a, is a family of processors designed to deliver the best price performance for your cloud workloads. Uh is what Amazon tells you. Uh, and, and so it, it also, uh, we, we tried to implement this across our ac two s. Um, we got a fair ways through it and then realized it wasn't gonna work because, um, one of our security agents didn't support arm based CPU. So we had to roll that back. Uh, it was a sad day for us all. But, um, uh, that's since been fixed and we're gonna, we're refocusing on that and in 2024 but one thing that you can do that's like totally brainless is just move your uh Aurora RDS clusters to Graviton. That's literally like a couple of lines of Terraform or, or a couple of clicks in the, in the console and you have um Graviton processors on your Aurora Postgres uh or MySQL and you're just instantly saving both cost and emissions. So, and that, that doesn't take uh any extra know how really uh unless you're doing something really complex probably. But uh if you uh were looking closely at the, the, the previous side with all the arrows, there was one place that did not have an arrow which was S3, which was a major uncovered cost.

So anybody who's doing data at scale is gonna know that S3 really creeps up on you. You think 0.002 0.023 cents per gig, this will never cost me any money. But when you're talking about 15 petabytes that ends up being a, a pretty significant sum of money. So this is something that we had to control for still. So this is a chart, this is a real chart of what our, our, our, our costs in S3 looked like. Um and uh the spoiler that's on the slide as we implemented Intelligent Tiering, which I would recommend everybody does. Uh but the orange slide was uh our, our linear growth with S3 and, and we had this plotted from way before this and it was right on the linear line which makes sense because the data we would get was pretty standard comes through just like we'd expect.

Um and so the the yellow line then is what happened when we implemented Intelligent Tiering. So S3, Intelligent Tiering for anybody who doesn't know is a uh uh S3 um storage mechanism that moves your data automatically from frequent access to infrequent access at 30 days of untouched to archive tier at 90 days of untouched. And I believe there's two more tiers beyond that, that are optional that you can move it to. Uh and in those first three tiers, if you touch or open a file, it will automatically come back to the standard tier uh and be fully available and then it'll start to count down all over again. So unless you have really random access S3 patterns, um this is uh a no brainer thing you can implement that will save you costs."

And so you can look at this chart here. Uh again, this is real, real. We were trending, right? Our costs were trending right with the the uh the storage line and then we implement an intelligent tier and you see the dip uh that's the first tier of uh files uh being moved off to the the 30 day archive. And then you see another enormous dip uh when things move off to 90 day archive.

And so this is the way we could have this data available uh for anybody that needs it because it's kind of a choose your own adventure in our environment. You know, the users can use it however they want. Um but if there are file files that aren't accessed that often they're available and just not costing us any money.

So then we ran this up against our uh carbon emission calculator and lost our minds like that, that bright yellow line is the s3 uh carbon emissions cost. So that lines up uh you can see that the bright yellow one drops literally off a cliff. Like you can see like we saw this and we are jaws to the floor because that's a crazy emission savings and we had a crazy cost savings by doing this.

So overlaying these looks like this. Um the the, the green on the back is the, it's the carbon emissions savings. You can see that it trends pretty well with. Um and we only had emissions uh data back to 2020 unfortunately, because that's when the calculator was implemented. But um you can see that we had a pretty clear trend line with what we were doing with cost. So this is the cost overlay with the emissions.

So all that to say, uh you're also not in it alone when you're with amazon here, right? So, uh we are doing these things to save it, but at the same time, amazon is becoming more efficient as well. So this is a map. This is uh I don't know how this is probably a few weeks ago map of uh amazon sites. Um and what kind of renewable energy they're making use of. So while you're also making changes to your workloads, uh amazon in the back end is also reducing emissions uh at scale by, you know, putting up solar panels and uh having wind farms.

So one other thing to talk about was those two spikes. I i keep going back to the data anomalies because i feel like that's the thing to talk about. Um so i left these in here again, real data. Um what those were is and this is our emissions. This is our total metric tons of emissions across these sources. Those two spikes were um us spinning up infrastructure in aws regions that were less carbon emission friendly than like us east one.

So we were doing a poc, we spun up some infrastructure, we looked at the emission calculator and we're like, what's this? Uh and so those spikes are, are, are we, we are learning that we need to be careful about what regions we're spinning up infrastructure into. So this is something that we now play into whenever we're talking about what to do with uh with our data. And where, so where are we now?

Um you can see here 2020 our estimated carbon emissions uh was 39.1 metric tons of carbon emissions. Uh which is uh uh i tried to equate this over to gas because i feel like people can, can conceive that better. Uh we started in 2020 at roughly uh 4400 gallons of gas used across our cloud. Uh and we've had a 91% reduction of, of that into 2023. Um just by implementing what i've just talked about, right?

So, uh that is uh wasn't incredible. So it's essentially taking us from like a family's uh a, a full family of uh three american families worth of driving to one american person's worth of driving, which is, is, was incredible.

Um so back to that original slide, this is stuff that it you could take away and do basically right now. So i'm gonna go through the, the biggest hitters is one is obviously moved to the cloud. Uh i realized that we're in a 300 level talk at aws reinventing and i'm like, hey guys, the cloud's pretty cool. And it's true though. It's really true. And it's something you can do right now with your organization, the more you can move there, there's other things that you gain by this, but also uh the the emissions uh are huge. And so you can, you can you get the economy of scale with amazon.

Uh and so i would highly recommend it. The next game changer for us was intelligent tearing. S3, intelligent tearing is almost totally brainless to implement. Uh and, and the cost and emission savings you saw earlier were uh uh demonstrable. So making that change is something you can do unless, unless you have very specific use cases with ss three, this is something you can literally go back and do right now. Uh and, and have, and, and be saving across the board so credible.

The the other stuff here has uh uh varying degrees of um considerations you have to make when you implement it. Um but the next two biggest for us was the um on demand where possible um which is the moving of your ec2 fleet to things like fargate and lambda. Um those can really change the face of your tech org and how you're, how you're implementing things and give you a lot of savings.

Um but they are also things that you need to consider more that, you know, you can't just move an ec2 to lambda and call it a day. There's, there's a little bit more to be involved in that. So uh that's something to get involved with your tech team and your org and your uh finance team.

Uh and then the other huge one for us was uh number seven that the uh utilizing amazon redshift spectrum uh to down slice your data clusters. The example i have for that is we had, we have a number of redshift data clusters. We, we use very heavily and we love um but at the time, we had something like a 55 node dc 28 xl uh which is a very expensive, very awesome uh red shift cluster. And because of spectrum, we were able to tear that down to 20 node. Uh and then later tear that down to an eight node r a three and move some of the data on to the red shift cluster rather than using spectrum.

So we ended up because of spectrum and that the, the technology changes with red shift, we were able to make a lot of new changes that uh that worked to save us a lot of money. Um and to save us a lot of emissions. Uh but then beyond that, then you have now red shift on demand. Uh that is that is coming up.

The other stuff is also very important, but those are ones that change for us. Uh but certainly uh 12 and three will save you a lot and potentially four. Thank you for your time. I'm gonna hand it over to mica now and I'll be available again afterwards for questions if anybody has it. I apologize for the fast speaking. It's done. Thanks john.

Hi, everyone. How are you today? Uh I can't really see you but show of hands if I can anyone have a chance to get down to the fear to see the darren aronofsky film? Yeah, a couple of you, i highly recommend it if you have a chance before you go home. It was spectacular. Just a beautiful film, a very immersive experience. Uh but it really got me thinking about how critical this discussion around sustainability is right now. We have to start thinking about how we can embed these discussions in pretty much everything that we do uh like every meeting. So uh yeah, definitely check that out. It's a great, great film.

Um so, hi, i'm i'm michael walter. I'm a senior solutions architect at aws. Uh and i'm what's known as an account essay. So i work with customers directly uh like john and anelle uh to help them utilize our services in, in any way that they kind of can kind of imagine. Uh and so i was really excited when john, when john and an elect came to us and started talking about sustainability.

Uh one reason was that i realized it was something i didn't know a ton about at the time

And then the other reason was that uh you know, i, i just realized how, how important it was to me, how, you know, how uh much of a big concern this is for all of us today. So uh I saw this opportunity where, you know, i could potentially help john and help anal elect to kind of start down this journey of sustainability.

Uh but also to learn something along the way and hopefully help others other essays, other account teams uh work with, with you the customer to, to start your own journey with sustainability.

So i'm not gonna take up a ton of time. We'll have a little bit of time at the end for, for q and a. Uh I just want to give you a few things that you can take home with you. Uh when you go back to your organizations to talk, talking with your leadership, uh a few best practices that we've learned along the way.

Uh that john definitely touched on as well as racks that you can use to sort of start that journey of sustainability. So we're gonna talk a bit about leadership. Uh I'm gonna touch on the well architected framework, uh something i call pathways towards sustainability. So uh the different ways you can kind of get started in parallel with one another uh to, to start uh you know, reducing your, your carbon emissions. And then how do you measure your success we'll talk a bit about that as well.

I think the number one thing we have to really think about is getting our leadership on board. Sustainability is one of those things that requires a culture shift in your organization. It's not something you can kind of just do on your own in your own little pocket of your company. You really need to get the entire organization on board.

So to do that, find a champion at your organization who can be that leader to set that north star for your organization. There's a couple of creative ways I've thought about, about how to do this. One of those is that in many medium to large organizations, you'll find other kinds of sustainability practices already in motion. So whether it's, you know, a recycling program, uh or figuring out ways to have a more green space to work in that kind of thing, these things are probably already going on within your organization.

So find those, those leaders of that, of those initiatives and add your conversation to it. How do we extend this to the cloud as well? Another thing you can do is take a look at the climate pledge. You don't have to sign it today. It's not required. Uh but go on the website, go to the climate pledge website and see what other companies have signed, right? See if there's other organizations that are in your same industry or vertical uh that you can kind of compare yourselves to and use a sort of a source of inspiration.

You know, if they're doing it, why, why can't we and you know, talk to your account team, talk to us. We at aws are here to help. So if you have questions, we can find specialists to answer those questions. Whether it's having to do with moving to graviton or understanding the carbon footprint tool. Come talk to us, we'll pull the right people together and help you sort of chart your path.

And the other main one, the, the one that john mentioned quite a bit is associate sustainability with cost optimization. It's really easy to sell cost optimization to your teams. No one's ever going to say we don't want to save money today. So start start there, start with say, how, how do we save money? How do we optimize our own cost and then add the sustainability uh aspect to it kind of together.

So i mentioned the well-architected framework. Does everybody know what the well-architected framework is? Have you all heard of it? I hope so. It's one of the most important documents that we have on our website. So if you haven't heard of it after this session, please go to the aws website and look it up. It's an incredibly powerful document.

It's organized into what we call pillars and one of those pillars has to do with sustainability. So it's a great way to tell yourselves when you're developing your workloads. Are we doing the right things? Are we thinking about all the different aspects around sustainability that we should be, you probably already are doing a few of those types of activities, but there might be areas that you could improve on or that you could kind of balance your, your attention towards. So definitely take a look at that document.

So these are some of the pathways towards sustainability. It's not that you have to do all of these things all at once. It's not that you have to do just one. It's really about finding something that works for you now.

So i mentioned building this culture, right? Building a new culture around sustainability. I like to think of it as building a culture of reduce and innovate these two aspects will really help you kind of build sustainability into the dna of your organization.

It's a reduction. We talked about it quite a bit already. That's turning the lights off, that's racks, shit's story of of uh closing the door on his way out. So do these kinds of things in the cloud as well. We have lots of tools to help you uh reduce your aws footprint.

Uh there's a w storage lens, this is a really powerful tool for looking at your footprint across s3. You can use this to find places where you can reduce duplicated data, all sorts of different things that will help you to minimize o over over uh resourcing on s3, right? Sizing tools, we have to make sure that you're optimized on ec2, turning off uh ec2 instances when they're not necessary. The same thing as turning off the light switch life cycle policies for archiving data using intelligent tearing, that sort of stuff.

Uh and schedulers making sure that you are able to uh turn down workloads when they're not needed. Maybe it's on the weekends or at night time. This is an incredibly powerful tool to help reduce emissions as well.

And innovation is really about using the most modern technology. Earlier this week, we had an announcement around graviton four and you might have noticed that one of the key aspects that adam highlighted in that talk was the reduction of carbon emissions with graviton four every time we launch a new version, a new iteration of those processors that gets better and better.

So always think about how do we continually update to the latest technology so that we're taking advantage of what's available to us.

So lastly measuring your success, right? It's really difficult to feel like you're making progress unless you have the tools available to you to make to to track your progress. The carbon footprint tool that john mentioned is in the console now you can use it today. It's really simple.

Uh it has an api as well so you can download the data and analyze it on your own. But you can also just look right in the console and take a take a look at where you are, use that as a baseline. It'll tell you the top spenders for emissions.

Uh and you can start focusing on how you can reduce those based on that data. It's also great like john showed in the in the charts to see that that data sort of fall off the cliff. So take a look at that, but there's other kinds of activities you might want to think about as well.

We have cost optimization workshops available. Uh so if you have an account team, you can start talking to them about doing those, there's ways to do them on your own as well. And these will help reduce your overall spend. But they'll also highlight places where you can think about carbon emission reduction, kudos dashboards or cloud intelligence dashboards or another thing that you can install in your aw s account.

It's relatively simple to do. You can do it on your own through some instructions we have on our website or work with your account team to get that going. And they'll give you all sorts of charts and graphs, all kinds of data ways to set kpis ways to track your progress both around cost and emissions and finally talk to your account team.

I can't, i think i've said that like four times now, please don't wait till the last minute, come contact us, talk to us and we'll help set you on the right path.

So thank you all. We have some time for q and a uh but that's it. Thanks for, thanks for your time.