Exiting the data center: A League of Legends and VALORANT story

I think we are ready to commence. All right. All right. So thank you for coming to our session. This is Games 304, Exiting the Data Center, The League of Legends and Valorant Story. And this has been the, we're at the end of a, a very long cloud migration and it's been so many lessons learned, a humongous effort. I'm really proud to be on stage with Riah Games to share with you what the lessons learned were, what how this all started from the beginning. We're going to tell it to you in a bit of a storytelling fashion.

I'm Adrian Baron. I'm one of the Technical Account Managers with AWS supporting Riot Games. I have our two Riot champions on stage here. A Infrastructure and Operations at Riot Games. I'm very power, Infrastructure Engineer on the Infrastructure team at Riot Games.

All right. So, let's see. Where is this gadget? I will get it started by having Rory, tell you about Riot Games, how they originated League of Legends and that hyper growth. And if you all are interested in the games a little bit about the architecture. He'll go into AD T A and I are gonna speak to you about the migrations, the evolution of the program that became the global data center decommission. And this was a program that ended up with us migrating into 31 different sites globally. So this is AWS regions, local zones, AWS Outposts. So humongous project. No, understating the level of effort that I took and we'll wrap it up with takeaways at the end and we'll have time for Q and A if you want, we'll come off stage and have some personal one on ones down here. So without further ado, all right, take it away.

Rory: Thanks. So I won't ask if you've heard of Riot Games before. That might be cheating a bit because we've been to Re:Invent or, and you may have caught Brent Rich's keynote or section during the keynote on Monday evening. But I thought it would ask maybe if, how many people may, maybe by a show of hands have played a Riot game in the last year. That's a lot most of the people. Um that's great to see. Maybe how many would have played Valorant and League of Legends. Basically, both, I can't tell we'll have a face off at the end. They play all the games.

So, that's great that you have that, that context of who Riot Games is. But I'll take it back to the beginning. Riot Games was founded in 2006. We've grown since then, obviously to now, more than 4000 Rioters and additional contractors and support staff across 20 offices worldwide. We started originally with League of Legends as our, our first game that was released and we've since expanded to the list of titles you can, can see on screen including some titles in the League IP and then some unique new IPs such as Valorant. We also partner with Riot Forge, which has arrangements with external partners who have produce comple experiences more than the long term long running life services games that Riot focuses on as well.

So I'm going to take you back in time to 2009 then and explain the launch of League of Legends. So for those who don't know what League of Legends is, it's a five v five multiplayer online battle arena game or a MOBA, where you group up with four of the people and take on another team of five to complete an objective to destroy the enemy's nexus. You choose your own unique champion that you will play from a list of, I think about 160 we have at the moment and take that onto the rift and, and battle with your friends and try and win the game.

It was extremely popular when we launched it and we quickly realized we needed to bring this to players all around the world. Riot is a very player focused company that's at the heart of everything we do. We want to make it better to be a player. So yeah, we set to this task of taking League of Legends to the entire world starting in 2009 and for the next couple of years.

To explain something about the architecture of League of Legends from the client perspective. So these are the screens that somebody would see when they're playing our game. On the left hand side, you can see that our platform experience. So our client connects to our platform servers and that handles things like friends list, matchmaking any of the progression systems, your unlocks. And that, so you can do all your out of game pieces there. Almost all of that is rest HTTP based over TCP. And it's not particularly latency sensitive. We could deploy these platform servers almost anywhere in the world and players wouldn't particularly notice the difference.

However, on the other, the right hand side, you can see the in game experience. So this is actually a separate binary that's run on, on PC. And that's what runs the in game experience when you're actually within the game, you've chosen your champion and you're ready to go and, and, and, and play for your 30 to 45 minutes game. So it's particularly latency sensitive. What happens is, as with many online games, your instructions, your mouse clicks, your spell casts are sent directly to the server which processes them and sends, takes the results of all of those inputs from all the players and returns them directly to your, your client. Any change to your, your latency between you and the game server has a massive difference in terms of your, your gameplay, your gameplay would feel sluggish if the latency increased or if your packets were delayed or discarded by any sort of internet weather.

So this is a key part of making the experience of League of Legends good for our players is to keep this, this latency for the game server piece as low as possible.

Now that you understand a little bit about the architecture of League of Legends and what it takes to run the two sides of our games, I can maybe explain, go to the next slide and talk about the data centers that we built to take League of Legends to the entire world.

Each of these red dots on the map is a data center that served League of Legends for many years. We also partnered with some external partners in Europe and particularly in Southeast Asia, with Gina, we had a long term partnership with them to try and accelerate the rollout of these, these data centers. It was obviously, I think everyone in the industry would know that building a data center is not a small task, requires many different levels of expertise in different areas and can take a long time.

So it took us maybe about four years between launching all the shards that we wanted worldwide partnering with Gina and also rebuilding some of our data centers consolidating them into larger ones in Europe that we launched in about 2014. So in that period of five years between launch and 2014, it was a very busy time for Riot. We continually refined our data center designs, kept on putting new, new investments in technology, new systems in there to improve the experience for players. But it was ultimately something that we, we struggled with because it took a long time when you need to plan out every single piece of physical equipment to support your games, it limits how fast you can, can move. And ultimately, it wasn't something that we, we wanted to be the best at.

We don't want to be the best data center builder in the world and we want to be the best games company in the world and, and focus on our players and the experience we can deliver to them. So once we had these data centers, I described before the platform system, another thing we did to try and accelerate how we would deploy services out to our players was that we invested in our cluster.

So our cluster is our name, internal name for the Riot container engine that we used to allow service teams to self deploy onto our own infrastructure. So much like we now have Kubernetes that is industry standard, we started with our own homegrown container scheduler on top of Docker and then migrated to Mesosphere DCS in the second version of our cluster that we call our Cluster Two. But what this allowed us to do is to move fast by taking the services that were required to support the current version of League of Legends and also expand it to give new features and new, new capabilities. They could all be deployed by the individual service teams on their own cadence on top of our cluster without the need for upfront planning and racking of hardware. And waiting for those long lead times in terms of power and space and cooling that we would need when focusing on the data center.

So this allowed us to deploy on both Riot infrastructure and partner infrastructure later when we would move to AWS. The next project in terms of accelerating this, this deployment of infrastructure, we decided we wanted to move to a, we knew we wanted at that stage to be a cloud first company. So we experimented with moving part of the League platform, the legacy monolith platform to the cloud. We wanted to bring down this long lead time, it took to, to deploy new infrastructure.

So we did that. We deployed our public BE environment to cloud in about 2016. And then also three of our European shards had their platforms deployed to cloud as well. We still kept some of our key services, our cluster, some of the databases, our firewalls still in the physical data center. So it was an experiment from that, that, that area we didn't go all in on the cloud just yet. But we also didn't realize the full value of what we could achieve in the cloud.

So what happened during that, that period was we, we almost treated it like another data center that these instances were statically provisioned. They, it wasn't very easy to scale them up and down. We hadn't invested in the automation to make that simple. We had basically taken the design for our, our data center and encoded that in Terraform so it could be managed, but it wasn't as easy as we as we would have liked it to be this, this had an impact because we had sized the shards quite large for the number of players there. So when the number of players just didn't meet that these shards became relatively more expensive. And, and not as good value for us, it also suffered from the fact that we were migrating to our cluster that I mentioned before. And that meant that we were now doing even less in the cloud. We were bringing things back into the data center, not intentionally but because of the migration to our cluster.

So this project had intended to continue and do all shards, but we ultimately paused it there in 2017 and didn't go any further. After this stage, things are, are relatively quiet. We have some platform servers in AWS. We also have some key services like publishing data where you were consumers of AWS, but it was now time to start putting our core game services into the cloud.

So we took this the opportunity of the release of Team Fight Tactics, which is a game mode within the League of Legends engine. So it uses all the same infrastructure that I showed you before. But as a strategy game within League of Legends, it's not particularly latency sensitive. So this allowed us to deploy to the closest region without materially impacting the latency or the player experience. They would just connect to the nearest cloud region when we wanted and get a similar experience.

So you can see here the game client, when connecting to our platform server based on the game mode would either connect to the on premises game server and be allocated a game there. Or for in the case of TFT, we would migrate that to AWS game servers and we would run them there.

Over the next couple of years, we continued this, this is a very successful launch for us. TFT itself is a very popular game mode within League and drives a lot of traffic. So we were able to in times of, of higher load during events or just events such as worlds that drive more player traffic, we spun up more and more game servers in for TFT in the cloud. We also took this time to invest in the automation behind deploying these, these game servers so that we could easily spin them up and down. We contributed to the Terraform to allocate dedicated hosts that was actually code written by one of the engineers on my team. And we, we also invested in getting the servers added to our Chef based automation system automatically so that we could literally increment the number in Terraform to deploy new game servers and then everything would be handled on the back end. And we use this, this for many years.

So that was great and we got to the end of 2019, but we were still a single game at that stage, still Riot Games. As Brent mentioned during the keynote, so all of this changed around the, the 10 year anniversary of League of Legends in October 2019. So we had always been working for many years on R&D titles. But at the end of this anniversary celebration, we spent the first part of it rightfully thanking our players for playing League of Legends for 10 years and being with us on that journey. But at the end of the video, we, we added a hype piece that I'll play for you here. And in this one clip that you'll see, we announced Wild Rift which had been talked about earlier in the video with a mobile version of League of Legends.

We announced the Team Fight Tactics would be available on mobile. So you could, you could play it on your phone or, or iPad.

We announced Valorant, a first person shooter that nobody had seen before that we wasn't, wasn't leaked and they didn't have an idea of it.

Legends of Runeterra, a collectible card game in the League IP.

Project L, a, uh still unreleased fighting game. Um and also in the League IP, an Arcane show on Netflix that we would work on.

Um so we had a lot to do, we had the backing of AWS with us. We knew that that would be our, our infrastructure provider of choice.

Um and uh yeah, going into 2020 we knew that the stakes were high. So I'll see if the actual video plays now and if not, we'll move on.

And finally we become Riot Games.

Um so going into 2020 that was our, our first launch was early 2020 with Legends of Runeterra the collectible card game in the League IP.

Um this was a cloud first launch. Um not being latency sensitive, that's more of a strategy game. We already had the entire game deployed on our cluster.

Um we could have deployed it to a data center but it was fully in cloud. Um both platform and game servers all in cloud.

Um this uh this use of our cluster really helped accelerate the game development and made it easy to, to uh to launch it in the cloud.

Um so that was our, our first launch went extremely well and gave us confidence for the new releases we would need to, to do going forward.

So next came the launch of Valorant, uh Valorant as a tactical first person shooter.

Um has a much, much more strict latency requirement. It really matters to our players, whether their, their shots hit, uh whether they can see the enemy before the enemy can see them and whether the game is fair, overall, competitive integrity is something we care a lot about.

Our players care a lot about and we wanted to deliver the best experience for them.

So in order to do this, we, we literally had to combine the forces of both Riot and AWS.

Uh we used 12 different regions. We had four AWS Outposts regions that were deployed in combination with five Riot data centers that were in locations that AWS just didn't have a presence yet.

And we needed to uh to serve our players in. We also used both edge networks, both Riot Direct who is Riot's own internal ISP.

Um which we use to, to peer with other ISPs close to players and take their traffic onto our networks as soon as possible.

Um in order to give them the best experience. So we used a combination of both Riot Direct and AWS Global Accelerator in order to uh to hit our latency targets and make sure we gave the best, the best experience to our players.

We've spoken about Riot Direct, um before at re:Invent and other events. So um yeah, I won't, I won't go into that too much here.

So then just to give you an idea on the top of uh what you saw before in terms of League of Legends, shard deployments, um you would now see a map here of the Valorant locations and see how we combined both the um the footprint and the Riot Games footprint in order to, to um hit as many of our players and, and, and give them the best experience we could, you can also imagine this then on top of the League of Legends uh deployments and, and we had quite a lot to, to manage.

Um however, so for League of Legends, we were then still in the data center. Um it was the one piece we had successfully completed the, the, the launch of these new games, which had gone extremely well, probably above everyone's expectations.

Um and they had done very well for the company. Um but we still had League of Legends almost entirely in the data center. Uh we now needed to do that to move to be a fully cloud first company and migrate League also out of the data center?

I'm going to hand it over to Adika who will talk you through how we actually got League of Legends and did that migration to get it out of the data center.

Thank you, Rory. Let's talk about migrations, right? Um but when I see migrations, do we really need to do it? Why do we need to do it? We're fine. Like we are running in the data centers, everything is, is going all right.

So um why do we need to take this risk? Um don't worry, we were at the same place where we were thinking the same back when we were doing this. So I'll take you over to the journey of, of what we had to do in order to get our services or games out of the data center. And, and what, how we made it to the final state.

If we put ourselves in the middle of a Summoner's Rift game right now, I would think that we are in the mid game state like we started off. Well, um we started with a high, we've played some bars that came to us. Ok. Now we understand our opposition. We also know where we need to get. We know the final destination. We know how we need to win. At this point. We just need to work together um try to find the right way, the right approach and then win the game. Um that, that's how I see it.

So, um at this point, what was Valorant and and Legends of Runeterra are doing was that they were cloud first. They were already in our cluster. They were doing all the good things.

On the other hand, we had League of Legends and TFT who were running a very big player base. And then we were having a hybrid kind of approach for League of Legends and TFT. We were in the data center. We were also some part of it. We were in the cloud, we had TFT game service running in the cloud service.

I think now time to clean up all the things and then try to see how we can get to the good state that we finally need to go.

Even before we even started working on migrations and cut overs and all that. The League engineering team, the tech foundations and the release management, folks, everyone they worked on how do we deploy League and how do we release the services that run on League called as a release train.

So they developed something called release train which actually maintains parity across all the shards. We have, it maintains all the versioning for how we deploy our services, for how we release our services. And it automates a lot of manual work that used to be done back in the day service.

Owners had to get on to this release train, had to on board the release train to release their services along with the patch that goes out. Um biweekly that you all know about um they had there were some strict regulations and rules.

If you need to get your service onto the release train, like you need to have like patch notes, you need to have all the configurations into one single mono repo so that it's easier for conflict management, um branching and all and all those services.

So you need to have like release notes, you need, your services should be able to like do a rolling restart, batch rolling restarts. For example, if your service was being developed or being deployed by any legacy system, you need to move out to the latest version of our cluster, which was running back then.

So all these things actually evolved, how we do CI/CD for League and it set up a good base for how the future of League will look like.

At this point, we must be thinking, ok. Now we have automated League of Legends, but there is no more manual work, click of a button deploy goes out. Um the release trend is actually helping us a lot. We are able to on board new services, existing services are all being deployed properly across all the regions. And also China, we must be ready for migrations, right? Not really.

Um so we if you recall the reason why we actually started to work on the release trend is so that it's easier for us to actually create new shards and then deploy new shards. So that that was the hope when, when all of this was starting and did it help? I think, yes, it definitely did.

So we were able to deploy all these new shards, create new shards very easily because we had all our configurations in one single mono repo. And then we were able to clone these new shards.

We had two dedicated teams to actually work on creating the shard automation um where they come up with all the services that are required for this new shot. To run every shard is different for us. Like every shard has a bit of different configuration, every shard is different, like a localization perspective, it's different.

So all the things lived in one place and then cloning the shard was easy deploying, it was easy and then it was really good for us.

On the other hand, we were also investing more in how we build the infrastructure as well. Like back in the day, like four or five years ago, developing all these base infrastructure, took us like months together.

First, we have to bring up the infrastructure, then the service teams comes on, they put their services, they configure their services. And then we talk about testing and deploying our older times. Like the the internal charges actually took us like about 10 weeks and then it took us like um couple of teams to do it and then a team and just like 10 people or so compared to the most recent, the internal shot that only took us like a couple of weeks with 2 to 3 engineers working on it.

Um and we wanted to go live with PB, like we wanted to rebuild PB with this release train and short automation in January of 2021. But we took our time, we had like run books, we, we did all the things we took our time. We wanted to make the automation ready so that we can go live because coming ahead we have a lot of data centers to migrate.

So we rebuilt PB in early 2022 and then um that has actually gave us more confidence. On the other hand, we were starting to hit um limitations in our software defined network layer.

So we wanted to find something that is more like cloud native tech stack approach and see if we can find anything more standardized. So we were looking at things and Kubernetes stood out.

Um let me see, show of hands who are nowadays are using working in Kubernetes every day. I see a lot of people. Exactly.

So um that was pretty obvious. So we were looking at Kubernetes which was actually becoming the industry standard at that point.

Um there was a good support in the community and then AWS EKS was offering as a managed service bundled with AWS support with all other AWS services. It was, it was going to be the obvious choice for us to integrate.

Um and that is actually now our cluster 3.0 so um our cluster two that Rory was talking about was now evolving into EKS Kubernetes based generalized platform for it was the go to place for deploying all the services at Riot game teams, internal teams, you name it.

Um RC three or our cluster three point was the way to go um separating um internal clients, separating the game teams um isolating them in these clusters is very important to us from a security standpoint and also from a noisy neighbors standpoint because all these games keep running like load tests constantly, they do development.

So we had to create like different VPCs for each game, for each cluster. And then we had like them isolated for like dev gets a cluster production gets a cluster, the load test gets a cluster.

So we were able to separate that and then we bundle that with the VPC peering along with some homegrown like deploy tools and we were able to easily migrate our customers from RC two to RC three without any down time.

And then the deploys were successful, we were able to manage all the security layer of it, all the security groups very easily.

Um so that was a good thing. Um like you can see that good elasticity that we had from AWS over here is that we could, we were able to create all these new cluster RC three clusters in region, local zone and Outpost, which was one of the very important step to us because we had Valorant already running in all these all these locations.

And also coming up, we had plans for League to have them in all those places. So it was very important to us that we find something that can be supported in all these three places.

At this point. I'm going to bring in Adrian for a brief amount of time. So that he can talk about how we look at this global data center decommissioning from an operations perspective.

Wait, what? Ok. So we're kind of getting towards the end game here but not yet.

Uh, few points that I wanted to bring up on how Riot Games really accelerated their project, their program, the global uh global data center decommission.

So uh, Slalom was brought in, this is an AWS professional services partner. They had uh excellence in program management. They were already on board helping uh with hands on the the shard automation work. They were very pivotable in helping plan these once the business had set the shard exit or shard move dates based on lease end dates of the data centers and so on.

Slalom ran with that and they helped bring everyone together.

Another key thing that Riot did that if anyone that has done a migration here before knows is very scary is they did a live roll back um with the p pe environment that Aditya had mentioned. So this is the the their public beta environment. This is not a like a quiet, you know, low use shard. This is where all the best players wanna go. This has the latest features, the latest patches. It's a very busy environment and Riot took the big risk and uh makes me nervous even to say it but they rolled it up into the cloud migrated it, then they rolled it back on prem bringing the data back. Wow, that's huge. Then they rolled it forward again.

This is really key to proving to yourself as a business that we can do it and it the business, the stakeholders at the top, the infrastructure people on the bottom, everyone had much more confidence that now with pb e done, we could do all the remaining shards globally, right? And that that risk um was removed.

Another item that was uh really pivotal. I want to have a shout out for the Riot live operations team is uh these are the, this is the team that handles uh incident management um and player coms during issues, but they also took it upon themselves to uh really drive and refine the cutover steps, the the shard migration steps.

And so each one that we did, uh they took the lessons learned, they saw where there was friction where there was gaps where people ran into issues and they drove those down and they burned them down and, and it really reduced the risk for subsequent migration. So awesome work by live operations.

This refinement, the constant improvement of the migration steps enabled the next set of shard migrations. And Roy over there is wearing the t shirt with all of them on the back. You'll see uh to get done quickly.

Here you go. Hadi let's wrap it up soon.

It takes, it takes a village to actually do all these migrations because it's not like one strike team that we can have, we cannot have like one team that can do all of this.

I want to give a special shout out to like all the teams that were enrolled in the league, all the teams on the League of Legends, all the teams on the player platform. We have major work being done by infrastructure and operations, our partners and then also our t over here who are with us in those war rooms.

Um be it night or be it day. And then the database folks, especially the database folks did over 200 database migrations in this whole project. And then there was amount of work and our project managers who actually glued us together because, you know, engineers have a tendency of drifting away from, from like the work and then they're like, ok, hey, hey buddy, what's this deadline that you got here?

So that was very much look at because we were talking about months of work here.

Um so that was being done there.

Um short migration started to become like a routine. So we had done pb and then we had done, we had planned, we had a big plan of all these um being coming up in 2022 and 2023.

Um we were building a run book, we were iterating on our run book. After each migration, we had like roll back plans. After each migration, we had retros in case what we need to do to improve. For next time, we were improving our downtime for players as well.

Pb actually, the rebuild took over eight hours of player down time compared to the most recent sharp migration just under three hours. No surprises, no major issues. Everyone who were involved in these shar migrations, they knew what they had to do, they knew what time they had to come in. What were the steps they need to do what configurations they have to apply and deploy.

We were having like those common culprits that usually come out what the issues are coming up. We knew the solutions for them. So we had everything planned, even people who were not involved with this migration from the very beginning, if we gave them the run book and they would be able to easily run with it.

Just the three of us who are like so deeply involved in these migrations can be standing here and talking in front of you and there could be a migration tomorrow happening, not just there is one happening tomorrow.

Um and, and then yes, my team is actually working right now on making sure that we get that done for players tomorrow. So that's what that's what's happening.

Gina, Gina, Gina was our long standing publishing partner in southeast asia for league and tft. Since 2012, our partnership, our contract with them was ending.

So we had to make the decision that we will self publish league and tft in southeast asia. So the question was asked to us, are you able to spin up five shards, put them live at the same time? So that players don't have any downtime between the contract and, and between riot takes over?

I think the obvious answer, you would guess the answer was yes, because we had all the tools, all the instrumentation, all the automation, everything was ready. By that time we had done pb, we had done few live migrations. So we had the confidence in doing it. And that's what we did.

We built five new league shots in southeast asia along with the game servers in order to not disrupt the player experience. What we did was we launched all these game servers in southeast asia in singapore region. And then we just to make sure that players don't get any downtime, they just get to play.

We had very much strong co-operation from gina so that we can have account linking, done so that players don't lose their cosmetics, their rank information, their player list, their friends list and everything.

So everything was ported over and players started to play happily. We did this in early 2023. At the same time, we were also talking with aws tams on like how can we improve the latency for a subset of region of players who are not having the best ping at that point from singapore.

So we started looking at outpost we started looking at local zones and then we started to place orders with them on how can we get this capacity? How can we get the hardware? We're talking about thousands of servers here. So it was, it was a lot of work going on parallel along with these migrations.

So eventually, we were able to move all the players in those regions to outposts and local zones. Now, they're able to play with great pain and latency in southeast asia.

While all of this is going on. Gina migrations, live migrations release train is deploying league every two weeks. That's not stopping, new content is being dropped.

We were also working on how can we optimize what we currently have in the cloud? We were looking at fleet refreshes and then we looked at c six instance type for league game servers and then it was a couple of months of big project to migrate all of the league game servers to c six i worldwide.

We also looked at what are the other things that we can do? Some microservices for league were in our cluster somewhere outside of our cluster living on two instances for them. We decided we should migrate them to m six. i give them beef up the instances and then give them more memory allocations, give them more cost efficient instances. And it was like a four month project. i was like a 3 to 4 engineers.

We had migrated the whole platform servers to m six i. The infrastructure and operations team are also working on how we can do like enable horizontal part auto scaling that was mainly used by valorant so that we can scale down when we don't need the capacity. And then also we can scale up when we actually need the capacity.

Um that hugely helped us bring down our costs. Few other improvements which do it like custom termination behavior in ads so that we can drain those pods gracefully so that we can scale down efficiently.

So this actually helped us improve the total cost of ownership on how we actually manage each service, each product team service. And if they're using all these services that we are providing them, um it, it, it did help.

I'm going to give it to Adrian now so that he'll just talk about some takeaways and then what did we get out of this?

Ok. That was a long road, almost done. So i hope that uh you, you've seen the beginning, the mid game, we're at the end game here. What are the takeaways?

Ryot really made a lot of big tech investments early on, not necessarily for the purpose of moving to the cloud, but they made a huge impact. So the terraform investment infrastructure as code skill set becoming really, really good at that. All their teams made a huge, huge impact in their ability to move quickly into cloud, uh our cluster and starting off with that mindset of in abstracting infrastructure from devs very important the uh elasticity of aws compute, turning out compute where you want it, turning it down where uh player demands change, getting the compute where you need it as well. If it's an outpost rack or server in uh a data center, they're far away from a region closer to your customers, closer to your, your, your players.

So we've come full circle. Can i see a show of hands anyone ever seen one of these in real life? I figured there would be a lot less. This is a uh unless you've shredded 20,000 hard drives. This is an enterprise humongous uh hard drive shutter.

So de comming the data centers don't want to discount that. That's a, that's a big process. But um uh now we're kind of at that stage where we're uh uh getting done with the project, still some work to do, but already realizing a lot of value a lot more than just getting the game servers, getting platform into the cloud.

This is uh you, you can't be understated that these are live game servers. The players need new content to get dropped, they need their bug fixes. Those release, that release train never stopped through all of these things. And the program evolved. It wasn't just moving the shards, it was optimizing, right?

We did a first move then optimize, we did the guina migrations throughout the whole thing, they kept going and kept their players happy player experience above everything. That's the right motto.

The other things that this unlocked from a business perspective was the gameplay modes uh used to be a little bit disparate all that uh shard uniformity that uh configuration management work that they did allowed same type of gameplay globally. So the players around the world didn't have to look on some geo and say, i wish i could play like them. It's the same everywhere.

So yes, shorter time to build charges. Getting that infra ready is now a lot easier.

Um the up time all these things, they don't have to worry about managing data centers, dealing with uh hardware refreshes yet other than terra form work.

So we've reached a lot more players and a lot more parts of the world by now as well and have a lot of scale also to address things like soaking ddos.

So what is this unlock for for riot games? Yes, you know, aws uh handles the blocking and tackling for infrastructure and and getting their compute and delivering managed services for them. And this allows a, a very strategic partnership to happen once that those tactical things are are handled.

So um this is so much more than just riot leaning on us as, as a a vendor, right? This is aaa true partnership.

Um we advocate for, for riot inside aws and um we we fought together so to speak in the valorant term uh to collaborate on, you know, where we're gonna put local zones, uh new features, whether it's udp changes and global accelerator for the valorant use case or nlb security groups, uh pod security groups for rrrc three. And this has been um such a a humongous project and excited to do more with, with riot.

So i wanted to uh wrap it up and say thank you to riot games for sharing their story. Ro adia, come on stage, come on chip javier and uh thank you guys for attending our session. If you liked it and you want us to come back again, please uh open up the survey app and uh give us a like and a sub.