Saving Up to 60% on Compute Costs Using Amazon EKS and Karpenter with Prodigy Education

Prodigy Education has been supporting educators and parents since 2011 by harnessing the power of game-based learning. As of 2024, the company provides game-based education to help students worldwide improve their math and English skills. Prodigy Education supports over 1 million educators across North America, has an average of 9 million students, and boasts over 1 billion educational questions answered each month.

With as many as 200,000 concurrent players on peak usage days, the gamified educational service needed to be both scalable and reliable. Prodigy Education’s previous solution did not consistently scale fast enough to match peak demand, which led the company to overprovision infrastructure to keep its games running smoothly.

Further, in the company’s previous solution, the infrastructure team at Prodigy Education had to create, update, and manage most of the infrastructure for its developers, which led to bottlenecks. The company wanted to provide autonomy for its developers so that they could change infrastructure on their own.

To boost developer productivity and achieve adequate scalability and reliability for its game-based solution, Prodigy Education opted to migrate to Amazon EKS in 2019.

Prodigy Education wanted the migration to Amazon EKS to be invisible to its users. To achieve this, it worked closely with its account team and specialists from AWS throughout the migration. “The AWS teams made sure to get questions answered for us quickly when we needed them, which helped the migration run smoothly,” says Adam Delyea, director of engineering for the infrastructure team at Prodigy Education.

To further support a seamless migration, Prodigy Education initially sent only small percentages of its traffic to Amazon EKS to build up developer confidence and confirm that the games would run reliably. When its internal testing and validation was complete, Prodigy Education fully migrated 30–40 of the services used to run its games to Amazon EKS with virtually zero downtime. “Migrating to Amazon EKS was a great choice, and we were able to complete it quickly,” says Alexandre Garcia, principal engineer for the infrastructure team at Prodigy Education.

Prodigy Education also created isolated sandbox environments for developer testing, which also run in Kubernetes. In early 2023, Prodigy Education began to use Karpenter, an open-source, flexible, high-performance Kubernetes cluster auto scaler built on AWS. “Using Karpenter as a Kubernetes manifest, we can experiment more quickly,” says Erik Krieg, staff site reliability engineer on the infrastructure team at Prodigy Education. “Karpenter is a fantastic contribution that AWS has made to Kubernetes.”

As of 2024, both the math and the English games at Prodigy Education run on Amazon EKS, and the company uses Karpenter to spin up and configure nodes automatically. Karpenter spins up requested nodes in less than 2 minutes, a significant improvement compared with the 7 minutes required by Prodigy Education’s previous solution. The company has multiple Kubernetes clusters, including one for main production with the rest for preproduction and internal tooling. Across these clusters, it has over 300 nodes running concurrently daily. This figure can rise by almost 50 percent on Fridays during peak usage. Using Karpenter and Amazon EKS, the company can scale quickly to accommodate usage from 2,000 users overnight to 30,000 users in the morning to over 200,000 users in the afternoon.

By migrating to Amazon EKS, Prodigy Education also achieved its goal of increasing autonomy for its team of developers. By creating sandbox environments that are similar to the production environment, developers can be confident that their services will run effectively in production. “Using Amazon EKS, we can provide tools to our developers so that they can spin up new services with little to no dependence on the infrastructure team,” says Garcia.

In addition to the scalability and productivity benefits, Prodigy Education has also optimized compute costs and increased its solution’s reliability. By making its compute usage more efficient by using Karpenter, Prodigy Education saves between 40 and 60 percent on compute costs. Additionally, Prodigy Education’s use of Amazon EKS and Karpenter have improved incident tracking and reduced the number of service events, which improves its solution’s reliability. “Using Amazon EKS, we have the tooling to perform fast rollbacks should an event occur,” says Krieg.

Since migrating to Amazon EKS and Karpenter, Prodigy Education can now spend more time innovating. Using AWS, Prodigy Education has access to the open-source community, which helps it to gain insights from other developers. “We’ve been able to get source code and ideas on how to solve problems,” says Andrew den Hertog, staff site reliability engineer on the infrastructure team at Prodigy Education. “There’s a lot of power in the open-source community.”

The company has also begun experimenting with large language models on AWS. It started by self-hosting the large language models to learn their intricacies. Using Karpenter, Prodigy Education can experiment quickly due to the auto scaler’s efficiency.

“Using Kubernetes services on AWS has been fantastic,” says Radu Trambitas, manager of the infrastructure team at Prodigy Education. “We have seen Amazon EKS progress and develop and have adopted the new tools as they were created.”