Migrating critical business applications to AWS Graviton with ease

Steve: Awesome. Well, uh we have an exciting session I think for you today. Super pleased to see all of you turn out. Uh my name is Steven Jones. Most people call me Steve. Um I'm the general manager of our SAP business at uh at AWS. And we've been doing a very fun project, a media project with SAP uh over the last couple of years and I'm joined today by Stan and Matt. I'll let them introduce themselves just a little bit.

Ok. Wonderful. Thanks a lot, Steve and uh great to see you all. Uh my name is Stefan Boyo. Uh I'm from SAP and I'm actually heading uh all database engineering, uh which also is including Hana and Hana cloud and all other database assets as well and uh hand it over to Matt to assist him.

Matt: Great. Thank you. Hello, everyone. I'm Matt Zenas. I run our global SAP Hana and database product management and strategy team. Uh so I'm actually in Stefan's Hana and database development organization, right?

Steve: Fantastic. All right, folks. All right. So a few things of what we're gonna cover today, but the really the goal is to uh help you as uh many of you are probably also underway or considering taking a journey of maybe porting some of your most critical workloads uh to Graviton as well. How many of you have heard of Graviton? Of course. Um and so the goal here is actually share some lessons learned from the project uh along the way. And so, uh that is the goal. Uh we're gonna take some questions at the end, so be kind of thinking of those as we're going along and, uh, we'll, we'll give you some time for that as well. I want to spend a, a lot of time on that. But, uh, this project is, uh, is built on the back of a very long partnership that we have with SAP. Uh I'm sure many of you have heard how at AWS we like to start with customer requirements and work backwards. Um our, our, our relationship and the work that we've done with SAP is no different, right? So, very early on into, um, this thing we like to call cloud. Now, uh, customers were asking us, what do you think about running SAP in the cloud? Right? When we started thinking about this, um, it was very, very early on. And so we, we started a partner with SAP on just what would it actually take to run a mission critical application like SAP in the public cloud, right?

Um and, uh along the way, you know, we learned a lot. And uh early customers uh began to adopt a platform and uh all of those learnings went right back to, you know, research development and improving the product. And uh you know, midway through this journey, um SAP introduced uh a very high performance uh database called Hana. It's an in memory database. And uh of course, this, this requires special compute instances. I think at the time, uh the the biggest compute instance that we had from a memory perspective, I think it had maybe a quarter terabyte of RAM, right? And uh we were considering at the time, well, how mu how much memory would a customer actually need? Maybe, maybe one never two could never be that big, right?

Um but we took a purposeful decision at that time to actually start building custom hardware uh with lots of memory to support this in memory database because this is where uh SAP customers were telling us they wanted to go if you fast forward to today uh earlier this week. Um we just announced a generation two of our high memory instances with up to 32 terabytes of RAM. I mean, who would have thought my, my son who is a gamer likes to tell me that he's got a very big desktop, right? That's got 16 gig of RAM. And uh and I turned around and tell him that we have servers that have 32 terabytes in them. But anyway, you can imagine there's a lot of work that goes on to, you know, engineering systems like this very complex. Um and along the way, uh you know, we built higher level services and been working with uh with SAP and they've been on a journey of them their own and they gonna tell you a little bit about what that's been like uh here in just a minute.

Um um but uh uh over the co over due course of time, you know, we, we got together on um this journey towards Graviton and a little bit about Graviton. I don't want to spend a lot of time on this, but if you look across the portfolio of AWS instances, um of course, the largest um set of instance, types of supported Intel from the very beginning, right? We slowly started to introduce um uh uh instances with AMD processors, uh which are also very performant. Um and then customers started asking us, have you guys ever considered uh building your own, you know, your own chips? Um you know, really pushing us on price performance. And uh of course, the, the the partners that we work with very closely, um you know, told us how complex it is and it is absolutely complex. Um and, but we continue to listen to customers and the reasons why. And uh so we started to partner with a little known company called Annapurna Labs based out of Israel.

Um this is back in the 2016, 2017 time frame. And back then, we were actually trying to um figure out how to take all of the typically expensive work that happens in the hypervisor layer and offload it into hardware because we, we were seeing, I would say a not insignificant tax or virtualization tax. And so um we wanted to uh be able to provide all of the compute power and memory cap uh memory capacity to customers and not have to reserve a bunch of that for hypervisor um processing or usage. And so uh we, we, we spun up this joint partnership with uh with Annapurna, we built these cards. Uh and then we thought to ourselves, these were based on arm, by the way, Graviton, they were really Graviton chips. And um we said, you know, we think this is, this is something we really ought to do. And so we, we, we set to work, it took us quite a long time. Uh but over the last, as you can see here, three years, uh we've been quickly iterating on uh these Graviton uh chips and instances you can see in 2018, we released our first Graviton instance uh with uh you know, completely um customized arm processor for AWS. We do all the design um uh and work with the manufacturers obviously, and then put them in our systems. And um and then uh we got to Graviton three and we started having conversations with uh with SAP.

But I think the question is why, why, why is this uh so compelling? Well, so we uh we generally, when we compare Graviton three to their, their counterpart instance in the same generation, we're seeing upwards of 25% price performance improvement. The best part though is uh that, that we're saving a ton of energy. And if you can think about just imagine all of the data centers and, and servers and racks and what it takes to actually power the cloud, right? There's a, there's a lot of power there and we, we want to be responsible and uh as we continue to grow, we, we uh generally at Amazon have uh have, have a commitment and a pledge to reduce and to get to carbon neutrality. And so Graviton is, is helping us get there. Uh a lot of our key internal services and I'll talk a little bit more about our managed services have actually gone to Graviton.

So i mentioned a little bit about um uh some of the nitro cards, right? Those are the hardware cards that had the early um Graviton processes in there. Uh in these Graviton instances, there's an entire system uh that is almost almost entirely hardware. Uh so obviously, it includes the Graviton processor, there's an onboard security chip. Um there's the cars i talked about and then we have a very, very lightweight pro um hypervisor and uh what i can say is that in benchmarks and we've done SAP benchmarks and we've benchmarked performance against bare metal and these virtualized uh ni instances, we generally see um performance um differences that are really neg neg negligible.

So let's get into the meat, i guess of the conversation, which is around, what does it actually take to move an application to arm essentially, right? And as you can imagine, there are all different types of applications out there and um the complexity is different depending on what uh what you're thinking about, right? So as i mentioned internally, we've done a lot of work to take our managed services already. Uh and port those, those uh engines, the software that powers them to be able to leverage Graviton, right? So that's essentially not a lot of effort for you. Uh it's mostly a console selection, right? Um and then uh for serverless again, that's pretty easy. We make those, we make that available for you to actually use as well.

Um we start to getting into a custom code that you've actually built over the years. We find that uh java python code is relatively easier. You'll find a lot of um uh open source packages. Um you can just essentially instead of having to recompile everything, you kind of pull in the the most recent updates and you'll, you'll have uh support for arm, but then we get into more involved, right? C++ c database engines and that, that requires some work. And so i'm gonna turn it over to Stephan to tell you how they approached um the workload at SAP. And you'll find that. I think it was not real simple task. Not really. No, thank you a lot.

Stefan: So talking about uh more involved tasks and you know, C++ code and what that means in terms of porting, uh we will talk a little more in the next couple of uh you know, slides and say the next half an hour or so, what that really meant. So we'll give you that insight. Um before we go there, um maybe a little introduction, what SAP does and what the SAP cloud does. Um who of you hasn't heard about SAP so far? Ok, awesome.

Um in a few very numbers, um SAP is a market leading company around all enterprise application software, right across like, you know, core ERP and surrounding processes across many industries. Um it's uh meanwhile, in the ballpark of 100 and 6000 employees uh globally worldwide spread around 100 development uh locations and as you can see as well uh with a very large uh partner ecosystem that are, you know, um you know, driving that uh into the market as well and extending the the processes as well.

Um if we look at it from a portfolio point of view um at the core of the enterprise application software systems is the, the ERP specifically also the cloud, ERP uh which is uh S4H um which is basically taking care of all of the, you know, core ERP processes and the fundamentals like, you know, financials and, and others. Um and that is uh on the other side, uh extended into multiple different industries and there are specifics uh and at the same time, surrounded by a bunch of applications that are line of business specific be that, you know, uh HR related or spent management and business networks, you know, supply chains and these kind of things uh also uh customer relationship management, right? And uh um for quite a while, there is also an uh uh ongoing effort to infuse AI into all of these processes and the mission critical things that are relevant for the, for our customers as well.

Um and also recently with all the, the movement on the AI space uh and the business AI activities also at the at SAP, there's more things that happen in terms of, you know, automation and infusing more and more intelligence.

Um and at the bottom, you see uh the business technology platform, which is basically our technology uh portfolio uh which is uh infusing, not basically all of the the technology assets and the needs that we use for our own applications, but that also our customers and partner use to build their application and their extensions in the SAP system as well and on, on the business technology platform, um to give you maybe a little idea of what that means in terms of, you know, overall market presence and impact.

Um the overall SAP customer base generates 87% of the global commerce worldwide, which is basically touching something like $46 trillion right? So which means, you know, a lot of the world economy is actually running and powered through SAP systems and applications, right? And basically all of the the big global companies out there have SAP footprint in some shape or form now zooming into the business technology platform piece and there is a couple of elements in there and I'll read through all of that

But just to give you an idea of, you know, that is basically the whole reach of technology that are needed from an enterprise application point of view, starting from application development and application servers, right, and development environment and all of that through automation integration data and analytics into the AI um pieces that are basically all of the ingredients that that we um and our customers and partners need at the end to build enterprise software, right?

Um and as we talk about SAP Cloud and the databases, we are actually looking at the data and analytics pillar as part of the business technology platform, right? So talking about SAP HANA Cloud, um let me start with a, a bit of a, a history on how that happened and what uh what that journey was over the last 13 plus years and Steve already um shared about it, you know, there's an own kind of history on, on what happened.

Um it goes back to um 2010 when SAP HANA um launched the first time and obviously there were a couple of years before uh in our development labs uh that, you know, we are all of, you know, the the the the heavy lifting in the work to to get the product ready um to to launch. Um and this was back in the day, the first in memory data platform on the market, right?

Um and ever since there was an evolution about, you know, adding more and more functionalities, you know, going from more analytical workloads to more transactional workloads and everything in between in terms of uh hybrid transaction analytics processing, adding more and more functionalities in not only relational uh things but also other engines.

Um I see a couple of steps on that journey. Um and uh the last kind of major, you know, new um change was then basically bringing all of these, you know, technology and assets into the cloud, right, which is then using AWS um as as a hyper scalar and the infrastructure underneath. But also from the from a database point of view, there were a lot of changes that we did uh in terms of transforming the database into a cloud native managed service that that we provide.

Um from my, from my HANA footprint perspective, I was talking about like, you know, the the overall touch on global economy from an SAP portfolio point of view. Uh maybe what is interesting is that if you look at it from through the lens of the database, um HANA and HANA Cloud in particular, um that is uh in the ballpark of 72 73,000 global customers worldwide that are running directly or indirectly uh on an SAP HANA HANA Cloud database, right?

Um and actually this is mission critical systems and processes that are actually running earning on the on these systems, right? And Steve talked about it earlier, right? I mean, some of these databases can become very big, right? Um not only a single note but also in a in a scale out way um to look a little more deeper into what happens behind the scenes of HANA Cloud, right? And there will be more details coming in in uh in subsequent slides um on on the applications that use HANA and HANA Cloud as a database.

Uh it's basically the entire portfolio of enterprise application needs, right? From analytical applications like also our own analytics portfolio into transactional application into everything that our customers and partners built on the business technology platform for their own purposes, right? Be that stand alone applications be that extensions or whatever the scenario is right? And obviously the spectrum in between, right?

And uh uh specifically uh in in latest information uh days, um we see a lot of things happening specifically in new scenarios as well that our partners and customers build around uh intelligent data applications that make use of the the power and functionality, right?

Talking about that one, HANA Cloud is not only a relational database as you might assume, but it's in fact a multimodal database, right, which means it has a violation of engine in memory column. Uh but also it comes with a spatial engine, a graph engine, a document store. And uh some of you might have seen uh the the the latest announcements uh um just a couple of weeks ago when we announced the uh the new vector engine which is specifically relevant in uh the the current ongoing dynamics around the generative AI and large language models.

Um there's also special functionalities built in in terms of advanced analytics um which means own machine learning and predictive algorithms that are part of the database and the data platform um that basically can be used to to run um scenarios on top of that, but also more uh information modeling, you know, semantical elements um for kind of analytical processing at the end of the day including multidimensional things.

And on the uh on the tier side, um uh there is if you look at HANA Cloud multiple tiers in terms of storage where you can keep your hot and relevant data in memory for high speed processing. But you have also other tiers down to object stores and, and data lakes um where you can, you know, keep, you know your large amounts of data connected through the the these layers right now.

Um why did we um start the journey um you know, two years ago um to port uh HANA Cloud to um to Graviton to AWS Graviton, right? And actually there were um three kind of main motivators behind that.

Um the first one is about what is the the innovation potential that, that we see behind that, right? Um if you see uh the history of SAP HANA, like the, the, the, the many years that I showed you on the prior slide, as you can imagine, there is a lot of work um in the code base to really uh um as a system level software that it is uh for really like, you know, performance to the, to the largest possible extent, right?

You know, very closely to the hardware uh and the the infrastructure that, that we see there, right? Um and the question was, you know, what is it that we can innovate close on that, you know, hardware software interface?

Then also if it comes to ARM based architectures specifically with AWS Graviton, right. The second element um specifically um as that matters, a lot uh for cloud services is around price performance. So what is it that we can uh uh achieve there in terms, you know, reducing cost on the one side, but also improving on the performance on the other side, right?

And you know, how does that, that metric uh play out, right, as a key uh quality on that side? And last but not least. Um and Steve talked about it in terms of energy efficiency. That's also a concern that we at SAP have. I mean, what is it that we can do to reduce the impact of our technologies onto the environment? Right?

So sustainability and energy efficiency was also a key argument why that movement onto AWS Graviton was a key um project that we were driving forward in the last two years now to give you a bit more insights and what exactly happened during that journey and what we did uh in terms of that porting and the results that we have seen, let me hand it over to Matt to run you through that one.

Thanks Stefan. I appreciate it. So, uh actually, you know, Stefan and Steve did a wonderful job teeing up and showing you the what and why of SAP HANA Cloud as well as AWS Graviton. That was a great way to kind of set the stage. And what I'm gonna be sharing with you uh over the next several slides is the how the where and the when in terms of porting SAP HANA Cloud to AWS Graviton and specifically what I'll be getting into are some of the details around um the work streams, how we kind of organize the project, some of the challenges that we had both technically and organizationally, uh we'll get into the project timelines.

Um how we've, how we've gone through the optimization process to, to optimize SAP HANA Cloud on AWS Graviton. And then also, and more importantly to show you the results of porting SAP HANA Cloud to AWS Graviton. And also what were some of the lessons learned that we have?

So to be sure, I hope it was clear when Stefan mentioned it, SAP HANA Cloud is a highly engineered database as a service for very extreme workloads. And so when you talk about taking everything that he had mentioned, all those different pieces of HANA Cloud and you wanna relat that essentially that's a nontrivial task. There's a lot of moving parts to do that.

Now, that doesn't mean that a that a a migration or reporting to a w AWS Graviton is, is difficult. It's just that from a HANA perspective and a HANA Cloud perspective, there were some unique challenges that we knew going into this uh that we just wanted to share with you.

First is a, as Stefan mentioned HANA Cloud is a fully managed cloud native multi tier multi cloud database as a service, right? It's very high performing uh does transactions and analytics. And to be even more precise, it is the de facto database as a service for SAP applications and services. So, and that's, that's really big, which means it's running really important workloads.

The other thing too that we'll get into more details about is that SAP HANA Cloud has a lot of different microservice or a lot of microservices. It just a, a sheer volume of them. So those needed to be dealt with as we made this port to AWS Graviton.

And as usual, there's not always technical issues that you have to deal with. There is sometimes organizational ones, right? Uh in this case, uh when you have a mission critical enterprise database as a service like SAP HANA Cloud, there's a lot of interdependencies within the organization, there's development teams. So there's various development teams, there's operational teams, there's infrastructure teams, there's security teams, you, you can imagine, right? And all you have to coordinate and communicate with all them. And so there's a lot of lessons learned there.

Now, in terms of adopting AWS Graviton for SAP HANA Cloud, uh what we're showing here are kind of what I would say are the main work streams, the key pillars of the project. And in, in terms of, from a product and service perspective, there's really actually three key parts to it that will go through. Um and then we'll get into some of the testing aspects. But when you break down the whole project and take kind of a 10,000 view, 10,000 ft view of what we were doing. These were the main work streams.

The first is our continuous integration and delivery build pipelines. Those needed to be uh ported to AWS Graviton. So this included adapting the build systems to consume Graviton virtual machines. It included uh building multi arc images as well.

Also, then there's the micro services and we'll, we'll spend more time in in additional detail on this, in particular what we had to do. Uh because I had mentioned before just about the the sheer volume that we had, but this was about adapting ARM ready workload, schedulers things along those lines, uh adapting the micro services to the AWS Graviton platform and also switching uh virtual machine pools and then you have the database engine itself in terms of the database, the the SAP HANA Cloud database engine.

So this required uh doing command set adaptations uh engine adap adaptations to chip structure and also general performance improvements. And we'll talk more about some of the optimizations and the approach that we took. And then the of course covering all of this for all these big rock items, the continuous integration and delivery pipelines, the micro services and the database containers.

Uh this also then required the testing right. So this this required enabling the test platform to consider the Graviton infrastructure. Uh this also considered the full end to end uh multi arc testing and also performance testing. And just to put it in perspective, there were thousands of unit and integration tests uh both end to end to, to cover all the different scenarios.

Now this is a a very, very high level view of the SAP HANA Cloud architecture, but I wanted to share it with you to kind of show a couple of pieces here. First is around the building blocks that when we were moving SAP HANA Cloud and all the components to the AWS Graviton infrastructure, what those building blocks were and also the the vast and variousness of the different microservices.

So at its, at its base, of course, you have the infrastructure as a service, uh you know, the ARM based AWS Graviton infrastructure and then building upon that the Gardener managed kernel services. Uh if you're not familiar with Gardener, Gardener is an SAP uh kernel project, an open source project uh that we contribute to.

And then of course, then we have the micro services that are on this that are, that are basically via the SAP common services infrastructure, that's how we deploy it. And then on top of all this, the SAP HANA Cloud database uh container containing the code line for the database.

What we show over on the right hand side over here are just kind of a sampling of the various micro services uh that we, that we have that manage the SAP HANA Cloud service. So these are things like um security monitoring, backup, um HADR, you know, infrastructure elasticity, you know, things like that. These are just just a few examples of, of some of the ones that we uh that we had and that we had to uh look at moving to the AWS Graviton infrastructure.

Now, when you look at a project timeline, this is obviously a very, very high level view. I think the way I would really just try to describe it simply is that uh this is a journey, not a destination, right? Uh there's multiple phases to this. Uh you know, we talked about timelines and when we got kicked off, but, but essentially, um as we're showing here, uh we kick this off in early uh 2022 what we'll call the ARM foundation phase.

So this was like, you know, getting the infrastructure ready, getting the teams up, you know, things along those lines and only about six months later, uh in late q 3 2022 we actually had our first initial port of the SAP HANA Cloud database uh that, that database engine that we talked about to uh the Graviton infrastructure.

Then about six months later, uh we had some selected uh micro services, those HANA Cloud microservices uh moved over on, on the Graviton infrastructure as well as those Cloud uh CI CD uh services as well that we moved over.

Then uh late last quarter of this year. Um we actually had our very first HANA Cloud workloads running um in the AWS Graviton infrastructure, which was great and we made an announcement about this a few weeks ago and you probably saw some announcements obviously at, at the keynotes this week going forward between now.

And as you can see here, mid 2025 you know, this is an exercise of, of refinement of improvement of of optimization, you know, things along those lines and you know, continuing to work together with AWS to do that.

So let's touch a little bit more about the microservices uh because I've mentioned that a few times now and just to put things in perspective. So this is kind of the overview of the, the different aspects of the microservice that we were dealing with um and how we collaborated and migrated and, and containerized them.

Essentially what we are dealing with to start with was about 250 micro services that were part of SAP HANA Cloud. And uh the services that were mandatory for Graviton support needed to be migrated. So what we did is um we actually with the SAP develop micro services, we jointly with AWS pro serve um migrated those over to the a to the AWS Graviton infrastructure.

And the good news is for many of the open source services. Uh there are armed versions that that are available in terms of a a migration perspective. Uh the enablement of the services on the Graviton three nodes was done via uh what I call smooth migration approach. Uh you know, not an abrupt approach from the intel to the Graviton chips.

Um and we did this over time. And basically what that meant is that since there was no hard migration, this helped prevent um instance, downtime and things along those lines. And from a containerization perspective, uh we had to do the Gardener adaptation...

so again, gartner is the, the kernis project that's part of saps uh kernis project. um so adaptations had to be performed. uh this included working closely with the uh the sap hanna engineering teams and the sap gardner teams to do this. but this included the enablement of gartner chute clusters, the release of uh arm based binaries and support for saps hardened container for linux distribution, the gartner linux.

so what did it do? what does it look like from a performance perspective? um this is actually the, the current state of performance uh that we can share with you today. so what we're showing here is the results of a of a benchmark of sap ho a cloud running on uh a graviton three system versus an intel based ice lake system as this was based on similar system sizes of about 64 virtual cp us and 256 gigabytes of memory.

now, in the chart, what you see here uh on the x axis are basically from zero to about 10,000. the different uh test scenarios that were run and along the y axis uh is the percentage improvement of running that particular query that performance test versus the ice lake. so above the x axis is, is positive means that we saw, you can see they're up to 50 there's uh percent, some are even getting close to 100 and 50%. and then below that not did not um outperform that.

but here's what we found is that uh there is actually a very big difference between the performance and the workload that we're running. so that was what we saw is like the workload class made a big difference. so in this case, the analytic workloads ran much better um in graviton. um whereas the transactional ones did not, and of course, those are the areas that we need to focus on jointly with, with aws to, to, to work on that for those potential improvements.

so what is our approach for optimizing sap ho a cloud on graviton? um so what we're laying out here is kind of the, the approach that we're taking in terms of the the different segments and how we're doing it first is just from an overall optimization perspective.

um obviously, we're looking at various things that we can do. not just one thing. uh some of this includes code vector, like uh like um neon and and scalable vector extensions, uh compiler upgrades, compiler options, uh local code changes things along those lines.

and in terms of just the identification of areas of optimization, uh we're using the hana performance regression testing tests uh both via via scenario or on a case by case basis. and we're doing this these regression tests in the hana continuous integration continuous development uh delivery pipelines. excuse me.

and from an analysis perspective, analysis has been done with um internal and external tools and and monitoring data call stack profiling uh down to the very detailed uh micro architectural level uh using the armed statistical profile profiling extension. and of course, um working very closely and collaborating with the uh aws and aural labs uh to further optimize ho a cloud uh for both current and upcoming graviton uh processor generations.

so, um what were our key learnings and takeaways uh and what worked well? and what, where did we struggle?

um so, first of all, from a a project perspective, uh the way we approach this was a scaled scrum approach to coordinate multiple teams. like i mentioned, there was a lot of different teams that we had to engage and work on this project. and so within each of these teams, sometimes we had permanent and sometimes we had temporary uh team members are a part of that that helped work on the burn down uh for the for the scrums. uh but what this did require it was a lot of coordinations across these teams. a lot of communication and speaking of communication, uh there was a lot of communication, of course, uh with executives across sap and, and aws. so this was a, a great forum to bring up either showstoppers or blockers or issues and get them out of the way, you know, very soon. so that, that was, that was really critical and worked really well.

also what worked well was the the adaptation of the architecture agnostic uh docker images. and also the fact that we were able to adapt around 100 50 micro services uh from, from the intel based uh platform to the aws graviton infrastructure. and then of course, enabling all the required continuous integration and delivery uh to achieve the essentially the same quality standard that we had on the x 86 platform.

so where did we struggle again? not so much from a technology perspective, just more of an approach.

um first of all, our initial report was done very early on prior to even having those build pipelines available. so because of that, there was some uh decreasing developer efficiency because we had really aggressive timelines. uh we actually ported uh a lot of the aspects, a lot of the database and things like that before, a lot of that had been done. i think you probably saw that on the timeline where we did the initial port and about six months later, then we had the c i cd pipelines, we had uh the micro services and things like that. so that, that caused us a bit of uh some struggling with some of the developer uh efficiency.

um and then also the there was effort in provisioning some external instances. um really that was more of an internal uh issue in terms of, you know, when a developer needed an instance to do some troubleshooting and analysis, to get access to that instance, sab has very, very stringent security and compliance regulations. and so that was more of an internal process thing that we needed to work out so they could get access to that.

so at the end of the day, uh what are the results achieved? um basically, uh it's fantastic news, i mean, 35% better price performance for those analytical workloads. we talked about um a 45% estimated reduction in the carbon footprint for sap honda cloud. and so what i can tell you is from an sap development perspective that um you know, the folks that were working on this project are really excited and proud of what they delivered and, and what they were able to do in that amount of time. uh obviously, there's more work to do and we're, you know, really looking forward to continuing to work with aws on continuing to optimize and refine and also innovate uh sap honda cloud on the aws graviton infrastructure with that, i think i'll turn it back over to, to steve.

awesome, thanks, man. so i don't know how many of you actually saw uh earlier this week. um we continue to obviously iterate on the graviton design and uh a lot of the, the learnings that um i would say many of their learnings that we uh we learned from the joint joint collaborative work here, uh which you can, you can tell was complex um have gone into uh i would say graviton four. and uh uh specific things like um well, actually, let's get a little bit into it, right.

so gravis on four instances, uh we released a preview of our first version which is rag instance, actually carry up to um 30% better performance. and that's largely driven by the fact that they have three times as many processors.

um and we're also uh delivering up to three x uh as the, the amount of ra m as well. um now, uh uh both stephan and uh matt talked a little bit about um uh performance of analytical workloads, right? and uh um some challenges with your transactional workloads. uh we're pretty excited about uh what we think um uh will happen with uh graviton four, especially for transactional workloads, given um uh advanced a lot of advancements in the prefect algorithms um and what's actually on on the chip itself.

so, um i don't know if you want to say anything about uh maybe some early work i can. yes. um, so we had a chance in the last couple of, uh, you know, days. uh, and, and, you know, a few weeks, um, as it's also shown here. right? i mean, it, it's in preview already. um, so we, we had a chance to, you know, look at the graviton four as well and run some of the, the, the benchmarks that you've seen on, you know, me, what, what you shared on the uh the performance test, right? and um there is, you know, very much excitement as well about, you know, what we have seen there, right? so we are seeing not exactly the 30% but really close to it. also on the analytical side in terms of, you know, improvements on three to current the four, right? uh and as you just said, right, you know, there's, you know, even more things that we see for the transactional workload, which is, you know, also very encouraging. so, um uh very good step forward on that side as well and, you know, uh looking forward to, to that one from a cloud perspective.

and there's the one thing is uh you know, the one is the performance side uh maybe to, to comment about. but uh the other one in terms of, you know, uh also the growing memory size and capacity that graviton four will bring us is obviously another uh kind of very interesting uh innovation and advancement right? because uh if you said it in the beginning, right, i mean, you know, for hanna specifically as it is in the memory database, uh quite a big, you know, demand in terms of, you know, uh memory capacity for for certain kind of scenarios and workload. so seeing like more memory and you know, bigger instances coming is for sure, something that is, you know, totally exciting.

yeah, i don't even remember at the beginning, i said we started early on in the journey of supporting hana with instances of a quarter terabyte of ra m. and in this case, gravitron three, i think, you know, we were working up into system sizes of half a terabyte. uh you can imagine with the r instances, we're gonna get up to a terabyte and a half of r am. and then we have a plan with graviton four to actually move into the x family which uh many, if you know where, where x is um there is multi there. so um it's not 32 terabytes of ra m i will be clear about this at the moment, but uh we love um i love it when customers push us. uh and we like to push boundaries as well. um and it's really um it's partnerships and, you know, key projects like this that uh help us push boundaries and really prove um what can be done uh as well as optimized, right to, to keep learning and and then everybody benefits from it.

um i wanna just uh uh just highlight a couple, couple things that are helpful to get started. um we, we put a lot of documentation together on how to get started. um and where, where to get help and how to actually sequence uh potentially uh you know, complex pro pro project like this.

um and uh we've also developed and released, i believe it was uh just a couple of months ago, what, what we call a porting assistant. uh this is freely available. you can download it in github. you can actually um uh work directly with your own source code. we're not gonna store anything, this is freely available for you to use. it's open source. um and it'll actually um help you some simulations and, and work as you try to understand um uh how best to take your code bases and move them through um through to arm and graviton.

um it does uh support uh c c++ i'm just reading here, fortran because i can't remember all of the languages off top of my head, java go, we've really um really meant to cover a broad spectrum of things.

and i know this slide is kind of funny because we're right at the end of uh re invent. but this is mainly for if you want to learn more about all of the other things that we're doing together with sap and what other customers are doing um all of these sessions, um you know, have been presented, i think for the most part, but can be, uh, can be watched later.

um and finally, there's always uh things that you can do, uh self paced self-paced labs, um uh certification trainings of that nature. and i think at this point, i would love to open it up for questions. thank you.