Riot Games Cuts $10M Annual Infrastructure Costs by Migrating to Amazon EKS

As a player-focused game studio with more than 180 million monthly active users, Riot focuses on creating large, long-lasting live service games and delivering remarkable experience for players globally. Riot has large player bases in North America, Europe, South Korea, Japan, China, and Southeast Asia. To meet the demand for low-latency gameplay, Riot initially rented space in data centers.

As Riot scaled its operations, it adopted Mesosphere DC/OS (Mesos) to orchestrate its containerized workloads. But when plans to sunset Mesos were announced, Riot needed to evaluate new solutions for its game server platform. The studio needed critical features such as global reach for low latency, high availability, efficient resource management, and automatic scaling, which became increasingly important as Riot expanded its player base and game portfolio.

In 2021, Riot began migrating to Amazon EKS. “Migrating to Amazon EKS has unlocked automatic scaling for us, which has been a big cost savings,” says Press. Using Amazon EKS, Riot built a centrally managed developer environment that abstracts the infrastructure for its game developers and creates greater consistency across the company.

By migrating to Amazon EKS and using Karpenter, an open-source node lifecycle manager for Kubernetes clusters, Riot unlocked efficient resource allocation and management in addition to automatic scaling capabilities, which resulted in cost savings of 10 million dollars annually. The studio no longer needs to manage licensing costs or dedicated host logic in its technology stack. To consistently deliver new features and meet customer needs, Riot modernized its platform using more granular application modularization and containerization, which increased the company’s efficiency in deploying customizable services for each customer. “We can allocate resources on demand,” says Lance Laursen, staff systems engineer at Riot. “We don’t have to order hardware to spin up new clusters on AWS; we can simply try things and iterate very quickly.”

Riot’s managed developer environment helps developers spin up compute, networking, and storage resources in accordance with governance policies without directly using the AWS console. The company has standardized 80 percent of its infrastructure into this centrally managed environment, cutting developer workload on infrastructure tasks by 40 percent. Developers can now onboard to the environment and deploy workloads in under 30 minutes, focusing more on game development.

“That abstraction helped us migrate as quickly as we did,” says Zach Koncir, tech lead for cloud and infrastructure at Riot. “We had a standard, consistent interface, so developers can target a new environment to deploy their pods to the Amazon EKS cloud environment instead of the data center. They didn’t have to do much beyond that to migrate.”

As it launched more games and modernized earlier titles on AWS, Riot also transitioned from a multi- to a single-tenant cluster model, creating isolated clusters for each game or use case. This isolation has reduced issue response time by 50 percent and improved capacity management. “This approach enhances isolation and simplifies the management of capacity and cost allocations,” says Press.

Riot automated much of the management and governance of its 246 clusters using Terraform, Karpenter, and other open-source technology. Automation has also increased Riot’s operational efficiency, reducing infrastructure provisioning time by 90 percent. With Kubernetes, Riot can also use the same infrastructure worldwide, including with the service provider that it uses to host its games in China.

Most importantly, players see less lag and jitter as Riot keeps latency low by using AWS Local Zones, a service to run applications on AWS infrastructure closer to end users and workloads. The company also uses AWS Outposts—a family of fully managed solutions delivering AWS infrastructure and services to virtually any on-premises or edge location—to place workloads closer to players. Thus, Riot meets its aggressive 35-millisecond-latency requirement of its service-level agreement for games such as VALORANT.

Riot has also expanded into underserved markets in South Africa, Southeast Asia, and South America, regions that previously required significant hardware investments. “Now that we have all this automation and the developer environment on AWS, it’s simple to get a new location that we can use to serve players,” says Press.

Now that Riot has migrated, modernized, automated, and standardized its infrastructure using Amazon EKS, it is looking to expand its developer environment to take on more features and use cases. Riot is also taking advantage of the flexibility to launch games more quickly using AWS infrastructure. For example, Riot launched a temporary game mode for League of Legends, called Swarm, as a treat for players. “In this new world where we can spin up infrastructure on AWS in many locations immediately, we can now consider things that previously were impossible,” says Press.