How to Build a Scalable Analytics and AI Foundation with S3

Speaker A: Uh, all right. Hello, everybody. This is Kate Straschne from Dedicated. We are live today with a very special episode with AWS's Paul Megan. He's the director of Product Management and Data and analytics at aws. Now, before I bring on Paul, I just want to set the stage for what we're going to talk about here. We're talking about how to build a scalable analytics and AI foundation with S3. Now, if you're not familiar with S3, I want to bring you back 20 years. March 14th, almost exactly 20 years ago, uh, 2006, when they announced Amazon S3, which is simple storage service, and it was essentially announced as the storage for the Internet. Now, we've gone, ah, a long way. It's 20 years later. So we are going to talk about the evolution. We're going to talk about how they're providing the foundation for AI, uh, workloads. And as part of that, we're also celebrating that Amazon S3 is 20 years old. As part of that, we're giving a Squishy. Well, a couple of squishies throughout the show. And the way to play is you just put a hashtag. So wherever you're joining us from, just type in AWS piday into the comments and you'll get one of these squishies. Now, before I tell you who this squishy is, I wanted to see how many of you actually know the name. So I'm going to remove the hashtag for a minute. Drop in the comments if you know the name of this specific AWS mascot. Good thing it's not on here. So that's great. So I'll see how many of you get that right. And I wanted to also very briefly show you that I've actually met this mascot in real life last year at Re Invent. Here you go. Okay. Well, that was fun. I actually asked them to pour data on me, so that was done on purpose. Now I'm going to go ahead and bring our very special guest Paul up on our stage here. Paul, ah, welcome to the dedicated show.

Speaker B: Thank you for having me, Kate. It's good to be here.

Speaker A: Absolutely. I also love that your background matches the color scheme. It's all on plan. You know, I'm thrilled to have this conversation. And you have a very interesting background. Now, you've been working with AWS since 1997. That's actually around the time that I moved to United States end, uh, of 96 from Tajikistan. But this story is all about you. So I want to hear sort of how did you start there and then take us through your journey and talk about your current role as well.

Speaker B: Sure. So I started Amazon in 97. Um, I was a college kid. I was not doing well in college and I needed a summer job. And I'm from Seattle and so I just got, uh, this summer job at a warehouse in the industrial district in Seattle, um, at a place called Amazon that no one had ever heard of. And you know, walked in the doors and it was just like, had this amazing vibe. Uh, you know, everyone was in a band, everyone was like working really hard, you know, um, just obsessed over, you know, you know, getting books to customers really quickly and just fell in love with the place that was Amazon's like, first big distribution center. Um, and so started working there. Uh, you know, eventually got into it. So, you know, I worked in the first, one of the first data centers as well. Uh, you know, in it kind of hung around until 2006. And then, uh, you know, Amazon still hadn't made a profit. It was probably about the exact same time that S3 was launching. So it was pre AWS, pre Kindle, all that stuff. And I kind of was like, ah, this Amazon thing isn't going anywhere. So I left and went back to college, spent 10 years out in industry working on storage, and then came back to AWS in 2018 and have been working on S3 ever since.

Speaker A: Amazing. So you sort of left for, I guess around the time, um, S3 launched, but then you came back.

Speaker B: I left on S3's birthday, the first.

Speaker A: Oh my God. Are you okay? Hold on. Are you so. Oh, I love that. Well, thank you so much for sharing the story. Now, you know, 20 years ago, S3 started with two actions. You had put and get. And maybe talk about that to my audience who are not familiar, but you put to store an object and you get to retrieve it later. Right. Talk about the evolution since then.

Speaker B: What's new post you put on the screen there for just a second. You see the first line in there talks about, uh, providing storage for the Internet. And that's really how we thought about that first use case. And some of the big first use cases for S3 were really about static images on websites. And that's what Amazon.com was used. What needed this huge unstructured data store for that was like very easy to put and take out of. Um, and that was really the first big use case for the service. Right. And, um, even today, of course, we have much more sophisticated CDNs and stuff like that that are Internet Facing. But even today, just a ton of the static content on the Internet is ultimately backed by S3. That was really the big first use case for the service. Um, quickly after that, folks realized just how useful it was to have an elastics store for things like backups and logs and stuff like that. Data sets that you didn't really want to think about storage for. Frankly that's a big workload that came on after, you know, soon after those initial storage for the Internet use cases. And then, you know, what we've seen is just over time, like as new technology waves have kind of rolled in, S3 has kind of been picked up and put to work for various waves. And so, um, after sort of these big static content use cases, customers really started getting into building kind of big data. And what's kind of funny is that almost to the day, Hadoop was invented about the same time as S3. Hadoop was also invented in 2006. And these two technologies kind of ran in parallel for about five years with Hadoop mostly being on prem. Uh, and then at some point customers made the connection that just how well these two technologies work together. Uh, and you kind of saw about five years into the Hadoop sort of history story, um, to see these big data analytics deployments roll out on top of S3. And that just ushered in a whole another use case for the service. And today one of our most predominant use cases today is about data analytics. And we have some huge data lakes running on S3 obviously, and we've come a long way since then. And you see history sort of repeating itself now with AI, with customers now starting to deploy AI applications on top of S3, doing training, more and more inference vector storage, all of that stuff. And so we kind of see like another wave starting up and now kind of like full steam ahead on data for AI as well. And so it's really those sort of like, you know, basic fundamental characteristics of S3 just has made it very, just uh, a nice storage layer for just a range of use cases. And customers have found those use cases and deployed on top of us. It's been great.

Speaker A: Yeah, you know, I wanted to mention some numbers. So today S3 stores over 500 trillion objects, which is crazy, and handles over 200 million requests per second. Now what is this actually talk about that scale? I mean it sounds huge. Um, I just want to hear your, how are customers actually using these workloads? What are they actually running?

Speaker B: It is huge. Uh, and it's a super interesting system to see from the inside. It's just like an amazing accomplishment of engineering. Right. And I say this as a product person, not as, not as an engineer. I'm fortunate enough to like get to kind of watch the engineering teams build on this thing and it's just, it's really, it's really cool to get a front row seat too. Um, and you know the answer to your question about like what, what are those 200 million requests a second? I mean it's just like it's such a range of applications that builders have stood up on top of of S3. I talked about some of the big major waves of applications, of Internet facing applications and backup and analytics and AI. Um, and those are sort of big major applications that take up a big chunk of that request load that we were talking about there. But really it's like everything that you could imagine is customers have, have built on top of S3. And that's kind of one of the cool things about working on the service is that when you talk to customers you kind of go from one to the next and it's just completely different apps that folks have stood up. Mhm. On top of it. And so as you know, coming from the data world, analytics makes up a huge chunk of that request. Traffic analytics is very sort of read heavy, read hungry, wants to get after data, uh, quite a lot. Uh, but you know, it's everything under the sun, it's contributing to that, to that request load for sure.

Speaker A: I can imagine. Just like Amazon in general has everything, so it makes sense. Um, checking in on some of the comments here, so I see folks are getting in with their AWS PI day. So yes, make sure those um, hashtags are coming in so you can get a chance to win a squishy. Yes. We've got some people saying that they know that this is Buckets. Yes, good job. The name is Buckets. There was uh, a comment here that I thought was interesting that S3 is officially old enough that there are junior development, uh, deploying to it who weren't alive when it launched. I did not need that to feel old. They feel old in general. So I'm like, how do you feel about that, Paul?

Speaker B: Uh, yeah, the joke that I make is often like times when you talk about the data volumes, it makes you feel old because you can say like, oh, I remember when whatever a megabyte was a lot or a terabyte was a lot and you deal with data at scale nowadays, it just uh, that makes you feel old just in itself.

Speaker A: Right, right, right, exactly. So um, you know Right now companies are dealing with lots of different data types, right? We've got some of, some of them are structured data, unstructured data, embeddings, vectors. So that poses a challenge because when you throw in lots of ingredients, uh, into a soup, um, how do you deal with all that? Right. So I wanted to ask, how is S3 helping with this challenge?

Speaker B: Well, I think that's where uh, the simplicity of the service kind of really helps because it really is a very basic building block that customers can use to roll into any type of application. I think that's been a big part of why it's been useful for so many developers is like the API is straightforward, it's generic, you know, it delivers on what you expect it to deliver on, providing, you know, elastic, durable performance storage. And it, you know, it often doesn't try and go further than that. It just kind of delivers on that and you know, really relies on builders to kind of, you know, build on top of it. And you know, and we've been fortunate to uh, kind of get them be incorporated into a bunch of different applications. Right? That said, you know, we're standing up more sort of primitives on top of that basic put and get API that you talked about at the beginning. Um, that kind of like aligned to some of the use cases that are most common with the service. Just to help folks, just to give folks a modern set of primitives, just to make it as easy as possible to interact with storage. Even in cases where the builder doesn't necessarily want to operate object by object, but may want to operate at different layers of abstraction. That's m. The path we've been on now for a couple of years is trying to um, just work backwards from the workloads and work backwards from what customers are trying to achieve. I think we've learned over the years that the right set of primitives, the right set of building blocks in storage is just incredibly powerful to help devs just go faster and get to the more valuable aspects of building applications rather than the toil of uh, dealing with storage.

Speaker A: Mhm. Right. If you could take that off of your client's hands that they don't have to deal with that they could focus on other things totally. Before, uh, my next question, I just want to remind folks, it's the hashtag is AWS PI Day and that's, you know, backstory. That's because S3 was born on March 14, which is PI day. For those who don't know, PI is 3.14. I don't remember the rest of it. But I, I do know it goes on forever. Um, so the, the question I'll ask right before we do that giveaway is from George, and he's asking what are. I guess let's, let's talk about Amazon S3's top three accomplishments. So what comes to mind in the, in the past 20 years, Paul? Uh, what are the top three accomplishments that you can think of?

Speaker B: The top three accomplishments, I would say that like, um, you know, the. So security is always, is the most important thing we do. It always will be. Right? And so, uh, and that's just like true across AWS, right? Um, something a little bit more specific to S3 is the focus on durability. Right. And so S3 is sort of famously 11 nines durable. Um, and you know that, that comes from sort of math that we're running behind the scenes on mean time to repair and how fast we're able to bring storage online and redundancy and these sorts of, these sorts of aspects of the system. Um, and sort of the, the way that we think about durability, the way we built durability into the store, into the storage, really is sort of unique and I think in an important contribution to, uh, tech. Right. And so I would say that just like what S3 has done on durability and really sort of setting the bar, uh, for data durability is a huge sort of accomplishment that I would cite. Uh, another thing is, um, just the amount of applications that are connected into S3 today, um, just means that whatever you want to accomplish, you can back it with S3. There are a bunch of vendors that have connected with S3 and customer applications connect to it. Just that the S3 API, almost as a standard, just that so many are connected into, is another really great thing about the community and about the service is just that there's such a broad range of stuff connected in. And then the last thing I would just say is that when I go to talk to customers, um, uh, uh, a lot of times what I hear is that we don't think about S3. It just works.

Speaker A: It's like the Internet. You don't think about it.

Speaker B: Yeah, it just happens. It just works. Uh, and I think that especially in a world where technology can be very complicated, technology can be very fiddly, as I'm sure everyone on this call can understand and relate to the fact that we get that feedback and customers say S3 just works, I think is a huge accomplishment. It's a huge contribution as well. Just to Help, uh, developers focus on what matters, help them focus on the cool stuff and not focus on configuring storage and thinking about hard drives and all that stuff.

Speaker A: I think that's my favorite accomplishment. And Maggie agrees. S3 just works. Yeah, the fact that people don't have to think about it, that's the biggest accomplishment. We're getting more questions, but as promised, I'm going to do the quick giveaway. The way that works is anyone who's entered the hashtag is in there and you will get another chance. So keep the hashtags coming into the comments. So congratulations to Jim Johnson.

Speaker B: Johnson.

Speaker A: Awesome. So I love, I love the little confetti there. Um, so yes, you're winning one of these. Congratulations for everybody else. Keep the hashtags coming. We'll do a few more giveaways throughout the session. Now I wanted to take the question here from Jerome. Without S3, do you think AI and big data would be where it is now? That's a good question.

Speaker B: Um, what I can tell you is that these teams would have had to spend a lot more time thinking about storage and you know, nothing's free. I'm sure there's a bunch of people on the call here who are part of dev teams and you know, whenever you have developers that have to go spend time on one thing, they're not spending time on another. Right. And so, um, that, I mean we can say that for certain that customers would have spent, had to invest a lot more time and effort in just sort of the undifferentiated aspects of setting up and managing storage that they just didn't have to think about or deal with. Um, I can also say that like, I mean, Kate, you mentioned some of the scale numbers at the beginning. Um, you know, we often talk about how S3 allows you to kind of use R scale to your advantage.

Speaker A: Mhm.

Speaker B: Uh, and that's also certainly true. And this has helped a bunch of customers to move faster over time. Right. Taking advantage of S3 scale to kind of inherit the reliability, inherit the ability to burst, and uh, increase the scale of their applications in step functions. Um, these sorts of sort of attributes that come from running at a certain scale are inherited, uh, to customers regardless of their size. M. Uh, I think that's in terms of whether AI and analytics would be where they are today. I think that's a big factor is that, you know, we've just, you, uh, know we spend a lot of time obsessing over the, you know, the lower levels of the stack. So, so that you don't have to. And, uh, you know, we, along the way, deliver a service that has certain characteristics that would just be like, really, really hard to build on your own just because, you know, they only come at a certain scale.

Speaker A: Yeah, I think it would definitely be very different, um, without.

Speaker B: It would be a different world, like, I mean, in the elasticity of S3, like, you know, because, like, that's another thing that's off the table for development teams. You don't need to worry about, like, okay, I have a petabyte. How do I get the next five petabytes? Like, when am I going to run out? How do I manage that? Sort of. We obsess over capacity management, obviously, in making sure that we have capacity to scale ahead of customers so that it feels, uh, it feels like that the system runs at unlimited scale, uh, to any given customer. Um, and we spend a ton of time and effort making sure that that is the feeling that customers get. Um, but it's hard and it takes a lot of. It takes a lot of work. Uh, so the fact that customers don't have to think about it is just really great.

Speaker A: It takes a huge load off knowing that they don't have to worry about getting more storage or scale or any of that. They just, they can rely fully on you not running out of space, technically. Um, okay, so I wanted to move the conversation. So we have S3 tables, we've got S3 vectors, and we've got S3 metadata. I think it'll be great if you just spend a minute setting the stage of how those three. Pun intended, S3, uh, how those three actually help with AI workloads. Like, what role do they play in the grander, like, vision of S3.

Speaker B: Sure. So I'll start with S3 tables. So I'll just kind of give you the origin story on this, on this feature. Um, so Apache Iceberg, invented a few years ago at Netflix. It's an open data format. It's an open table format for tabular data, really, invented for data in S3, Parquet data in particular. Um, effectively, for those of you who aren't familiar with, with Iceberg, what it effectively is, is it allows you to. It allows you to find logical tables based on objects in S3. So it establishes this metadata layer that is as a standard that allows you to define these logical tables that then you can go query with tools like Amazon, Athena, emr, a bunch of stuff. Right. Um, and so this was invented, I think, 2017, something like that, at Netflix. And, and, uh, just has really Picked up momentum over the last several years and is now uh, emerging as one of the most important technology waves in the analytics space in particular. And customers love it. It gives a bunch of great features for managing tabular data, for querying. These tables that Live In S3 allows customers to decouple, compute and storage in interesting ways. We're seeing more and more of these big iceberg deployments. Um, now, uh, as customers have done this, what they found is that um, introducing this metadata layer adds some additional undifferentiated work. You have to maintain the tables, you have to compact them, you have to do orphan runs, uh, when you uh, replicate data, for example, or do other storage management activities, you have to be aware of the table constructs, uh, as well, sort of, sort of the way you manage the storage needs to be table aware. Uh, and so, you know, as customers sort of more and more started deploying Iceberg, you know, they started to call on us, uh, to ask about, okay, like I love iceberg, it's great. Um, you know, how can I, you know, run intelligent tiering in an effective way? How can I replicate my tables in, you know, in ways that respect the sort of the boundaries of the table and are catalog consistent and so on. Um, and so these are the questions that we were fielding from customers that got us down this path of thinking about establishing iceberg tables as sort of primitives in AWS in S3. Right. And so that's the, that's effectively what S3 is. S3 Tables is at its most kind of basic level, it allows you to create iceberg tables and they are first class AWS resources.

Speaker A: Mhm.

Speaker B: They have an AWS resource name and arn. Right. Which means you can refer to them with AWS services. You can do lots of stuff, apply lots of tools within AWS that assume um, that the resource has an arn and is a first class AWS resource. Right.

Speaker A: Okay. So those are tables.

Speaker B: Yeah. And then we also then can do interesting things in the background, like run compaction for you in the background without you having to worry about it, do orphan runs, the kind of that basic maintenance stuff. And so that, that's kind of what tables is kind of a response to, uh, customers who are running iceberg at scale and same story, like want to offload a bunch of the undifferentiated work to us, uh, they can focus on their app. Now S3 metadata, uh, is like S3 Table's little cousin in that it's based on S3 Tables. What S3 metadata does is it sits on the data path, on S3's data path. And every object that comes into S3, uh, it grabs all the metadata off of that object and puts it in an iceberg table that's managed by us. Everything that S3 knows about an object, uh, goes into those managed iceberg tables. And what it gives you is it allows you uh, on a SQL prompt using Athena or any other tool that can query an iceberg table, allows you to go in and just find out information about your data using SQL. So you can, like, how much data is in any given storage class. Where are my deletes coming from? Um, how old is my data? Uh, any question that you could think of now could be answered on a SQL prompt. And nowadays given where we are with natural language to SQL, it's just very easy to get into position where you can just now ask questions about the contents of your bucket via uh, a metadata table. That's what S3 metadata is all about. It's all about making it super easy for customers to understand what they have, how it's changing, that sort of stuff.

Speaker A: Okay, so we got the little cousin

Speaker B: covered that's tables and S3 meta. They're both iceberg based. One of them, again metadata is like S3 managed tables that we produce on your behalf so that you can get into that position of easily asking questions about your data. Now vectors, uh, has a similar origin story to S3 in that there's this new data type which is vector embeddings represent almost the language of AI. It powers semantic search. They power techniques like rag, uh, as customers have stood up agents and other AI driven workflows, these vector embeddings get generated and they need to land in a vector database somewhere. Um, again customers started coming to us and asking about it, about storage for these vector embeddings. Um, you know, they become critical to the application as soon as I sort of introduce AI into the mix. Uh, and additionally, you know, customers kind of want this similar characteristics that they have from S3, right? They want the kind of low cost, very elastic, very durable storage for their vectors. And so that's sort of the feedback that we're starting to hear in this space and went down the same path that we went down with S3 tables. Um, instead of creating iceberg tables as AWS primitives, we started uh, creating vector indexes as AWS primitive, as an AWS primitive, uh, so that customers can just land their vectors, uh, in what we call a vector bucket in S3 and they land in an index and Then can be queried, um, which is in very similar sort of ways that you would expect in querying a vector database. And again in the background we're doing compaction and managing those indexes and doing all of that interesting stuff, uh, so that customers can just very easily get started storing vectors and building knowledge bases, running hybrid search. All these kind of cool techniques that customers want to do now in order to stand up AI in their environment. This just kind of takes storage off the table for getting started with those use cases.

Speaker A: Okay. And I actually have a follow up question around the vector. So for someone who you mentioned rag, so retrieval, augmented generation application or semantic search systems, why does it matter that that data sort of lives natively in S3 rather than in a separate vector database? What are the benefits?

Speaker B: Right. So you know, for, I'll say for personally and for my team, right. Like the, the biggest benefit out of the chute. I'll give you two out of the chute. What's really great about S3 vectors is that you know, just like S, uh3 it uh, scales from absolute zero. And so if you want to go experiment with AI, um, you can set up a vector bucket for nothing, put in a few vectors at a very low cost, stand up a bedrock knowledge base for very little and just get going. Running retrieval command, running retrieval calls, Tinkering with rag Tinkering with knowledge bases. When we put the preview out last July, that's what we saw. Just so many developers, you know, recognize this as just a very easy, low cost way to start to experiment with, with AI. And so like that's uh, the first thing is that just really brings down that sort of cost to entry to start experimenting with RAG and knowledge bases. It's definitely, it's how I learned about how to sort of play with this stuff. And you know, you know we see on my team for example like knowledge bases kind of springing up right, right and left because it's like, it's just such an easy thing to do and you don't have to think about it. So that's sort of like the kind of like the developer experience sort of getting started experimentation angle on y s 3 vectors m 1 thing that's nice about that is if one of those experiments takes off, um, it scales like S3 and so you're kind of fine.

Speaker A: Right.

Speaker B: Like you don't need to then worry about like migrating into some other storage. Right, Right. So uh, and there's this profile of AI applications or agents that just fits really well with the characteristics that you get from S3 just like, and this is true across all application spaces. Like there's a category of application that is fine with the kind of latency that you're going to get from S3. It's not like block storage, microsecond latency. And it doesn't pretend to be right. It's measured in the tens of milliseconds. Um, and there's just a category of applications in the AI space that like that latency and are perfectly willing to go trade off latency for cost, um, in order to sort of build on that cost profile. Cost. The price performance profile that you get from S3 generally in the vector. Does that make sense? So there's sort of like those two things just like easy to get started and experiment and then just like there's just a category of agents that are agent of agentic applications that likes that price performance.

Speaker A: Yeah, of course. Especially that first benefit. Because things are moving so fast. I feel like any friction you can remove from folks to just get started is already a huge benefit.

Speaker B: Especially in this space that's moving so fast. Like there's no, there's like no replacement for just playing with it and tinkering with it.

Speaker A: Yeah.

Speaker B: And m learning by experimenting. Right. And that's what we found on my team. You, uh, know, for I can tell you that that's just like the best way to learn is to just go play with it.

Speaker A: Get hands on. Absolutely. I completely agree with you. Um, we've got a comment here from a LinkedIn user. So I, I don't know who this is from, but uh, they say that. I think the magic of S3 is that the service is always available even as we upgrade, change the backend hardware and customers are free from worrying about aging hardware in their data centers. So do you agree with this? Do you think your customers are seeing this as one of the benefits?

Speaker B: Absolutely. Like this, this is, you know, a huge part of what we do. Right. And like we talk about it just works. Right. Um, so much engineering goes into making S3 what it is and a huge portion of that is doing stuff like this, like swapping in the underlying systems as they age, managing that whole aspect of storage. Um, we spend so much time and effort on that. Uh, and it's a huge difference from running on premises where you're always on this treadmill of depreciation cycles and getting arrays in and out of the building. It's like a huge amount of work and you can't mess it up. Right. And so for sure that Aspect has been a big part of what we do and why customers have deployed onto us.

Speaker A: Right. And I also wanted to talk about AI workloads. And I know you mentioned that it helps facilitate and removes the friction. But if you were talking to someone who started maybe five years ago, like a developer who used S3, what can they do now that they couldn't do maybe four or five years ago? When it comes to AI workloads?

Speaker B: Yeah. Uh, so when it comes to AI workloads, I would say the beauty of. Not the beauty, but one thing that is true about sort of like the Vector Store that we have, for example, is it is built on the same building blocks that you all have access to. And so it's kind of a trick question, like, what five years ago? What could a customer do now that they couldn't do five years ago? It's almost a trick question in that, like, we're using the same primitives that you have. And so, you know, you could go and, you know, build all of this yourself as well. Right. Like, you could go, you know, figure out all the knobs and dials and configuration, write the software and all of that stuff.

Speaker A: Who wants to do that? Paul? Uh, no.

Speaker B: Right. And so, you know, but. But it's sort of a truism of like, what we're, you know, kind of how the system is built. Right. And that. And so, you know, there's sort of, you know, you know, what we provided at least over the course of the last several years is like, just like, huge shortcuts to, um, offloading as much of the sort of undifferentiated portions of that that work, uh, off to us. Right. And so that's really the big, kind of, like, I would say that, you know, what could you do today that you can't do five years ago is like so much more because you're focused on what matters instead of the knobs and dials and, you know, the infrastructure, uh, that we obsess over when we come into the office every day.

Speaker A: Right. I love that. Thank you. Okay, we're going to do the giveaway again soon, so make sure to put the hashtag. But there was a question from Sean. Uh, what is your favorite overall launch and your unsung Hero launch?

Speaker B: Um, my favorite overall launch. So, you know, I have a few. I have a few favorite launches. They're all my children. I love them all equally. Um, uh, I think a big one for us. That is a favorite of mine that we did. I think 2024 was adding conditional, uh, rights to S3. Uh, this is like a very in the weeds change to the S3 API. Seems like it, yeah, yeah, write conditions. Uh, but it was something that developers had wanted for a long time and we just got such an amazing response from developers. Uh, and the adoption of conditional rights was just a rocket ship. Uh, I think it's like 2% or something like that of writes today in S3 are conditional rights. Which doesn't sound like much but given that those request numbers that you just called out Kate to get to 2% of a number that big so quickly is just very satisfying. And it was just a great reminder of who the real audience of S3Is. These developers are the real audience. And so that was a great. I don't know if that was a favorite or an unsung hero launch. Um, and then of course uh, uh, another big one is just that I think is like continues to kind of make gravy for customers is intelligent tiering which uh, launched Now I think 2019. Uh, this has been one of our most successful launches that I've seen in the service. Intelligent tiering, uh, automatically tiers your data down as it ages and gets cold down to cheaper tiers of storage. Uh, again taking sort of undifferentiated work off the table for you. And the benefit is it's sort of a save me money button right. And just really just drives down storage costs without customers have to think about it and I forget what the number is and just how many billions of dollars it saved customers or you know since launch I think we're above 5, $6 billion at this point that customers have saved by just wow, opting into intelligent tiering. And so that's something that you know, uh, it's just great to. We just love taking that undifferentiated work off of, off of customers plates. That's a big sort of a big, a big deal for us and uh, we focus on it a lot. So you know, when you, when you see the numbers and you see the money saved, it very gratifying. Lets you know that we've done something good there.

Speaker A: It seems like time saved as well. If that was going to be someone's job to remove all data or tier differently or treat it differently and you

Speaker B: do that analysis to identify the stuff that's not me. Right. There's a whole spreadsheet operation that needs to go into the front end of that that is just offloaded to S3 which is great.

Speaker A: Uh, we had a question here from, from Jay when you were talking about the knobs, that they don't have to turn the knobs, but did the knobs go to 11? I'm assuming this has. This is the 9 11s. Uh, I'm assuming it's a joke, Jay. That's the knobs go to 11. Um, Jay actually had another comment here about making two phase commit like this happen in S3 buckets is magical and how concurrency is the new currency. Any, any thoughts on that?

Speaker B: I mean, you know, concurrency is. It's been an interesting thing for me personally coming from sort of the on prem storage world and you know, the late teens to a system that is so distributed, so concurrent. Um, and you know, uh, like we. It's a whole separate way of thinking in terms of how to build applications to build for concurrency. Um, uh, and so I don't know, there's a lot written about that, uh, in terms of how the system is designed. Uh, I don't know if, um, Jay is referring to sort of the strong consistency that we introduced in the system a few years ago to simplify, to eliminate sort of the eventual consistency, uh, characteristic that S3 had for many years. But, uh, those two sorts of things, like highly concurrent, highly distributed system along with sort of strict consistency was like a really interesting engineering challenge that the teams overcame after a bunch of years and so characteristic in the service.

Speaker A: Now for sure he said 100%, so he's agreeing with. Um. So you've got the, the nail on the head. Um. All right, it's time for another giveaway. You ready, Paul?

Speaker B: Ready.

Speaker A: I'm gonna spin the dial here. Okay, here we go. We're gonna go ahead and draw again for one of these cool S3 squishies. And I don't think you can win twice. All right, Maggie. Congratulations, Maggie. Oh, you're, you're. You get a bucket and you get a bucket. Oh, buckets. So I, um, know we're getting more comments and questions. I know. And we also have to. I just realized we're 40 minutes into this. Time flies when you're having fun. Paul, just a few quick questions here. You know, I want to talk about what future problem you think AWS or Amazon S3 is going to solve next. You've come such a long way. Like, besides allowing buckets to get a drink on their 21st birthday, what else can we expect?

Speaker B: Uh, look, 90% of what we build comes directly from customers. And that's one of the cool things about working here, is that we don't overthink strategy and vision. It's really Our customer's job to think about strategy and vision. Um, obviously we spend some time on that. We spend some time trying to look around corners for customers. And there are some features that come out of that thinking. But when you ask kind of what's next, really, that's a question for the people on the call, not for me in a lot of ways, because that's kind of how we think about the roadmap. And we make the roadmap in calls and in conference rooms where we're asking questions and learning from what customers are building. And so, um, I can tell you that we're going to stay focused on scaling ahead and staying out of your way. We're going to hold that bar as security being the most important thing that we do. Um, we have these new primitives in the service around tables and vectors, and you're going to see those mature and scale and, um, get additional functionality as you guys tell us what, what you need. And, um, you know, that's the.

Speaker A: I love this answer because it just shows how much of a great partner Amazon is, that your future plans depend on what your customers actually need. And I also love how close you are to your developers because you seem genuinely excited about conditional rights and like, not everyone is that excited about, um, new little features, right, that potentially change, uh, lives and, yes, save money and save time. Um, but it definitely comes through your excitement and partnering. Now, the final last question I have for you is for my audience who maybe want to learn more about S3 in general. The evolution, about the metadata, the vectors, um, where do you suggest they go?

Speaker B: So, I mean, the S3 product page, we obsess over that too. We keep it up to date. Hopefully it gives folks all the information they need to get to get started. Um, so, I mean, You Google Amazon S3, you're going to end up on the, on the main product page and, you know, it should take you to. Everything we talked about today is out there. There's no, you know, you know, fancy futuristic roadmap stuff that we're talking about. So any you're interested in, any of it, you should be able to go to the product page, go to the FAQs, get pointers to the documentation and all that good stuff.

Speaker A: Okay, perfect. I'll share the link to the S3 product page in the comments so people can just click if they don't want to Google. Um, but yes, Paul, thank you so much for your time here today. I learned a great deal about S3 Evolution and I'm excited for the future and I love how, again, how you're partnering with your developers and your customers in driving the future.

Speaker B: Thank you so much, Kate. I really appreciate you having us.

Speaker A: Awesome. All right, say say bye to Buckets and thank you all for playing the giveaway game. We'll be in touch with you to get your squishy. All right.