Momenta Migrates to Amazon EKS for Automatic Cluster Scalability and Upgrades

Momenta, a Beijing-based technology company, uses machine learning to create the “brain” for autonomous driving. With a mission of improving vehicle safety and efficiency, Momenta is working to enable autonomous driving at scale. The company currently offers software to original equipment manufacturers (OEMs) that supports a variety of driving scenarios, including intelligent valet parking, automatic lane changing, and distraction and drowsiness detection. Since its founding in 2016, Momenta has been using a variety of Amazon Web Services (AWS) solutions, taking advantage of the AWS global infrastructure and AWS Activate startup support. In 2020, Momenta migrated its Kubernetes clusters to Amazon Elastic Kubernetes Service (Amazon EKS).

Momenta initially decided to use self-managed, open-source Kubernetes clusters for running its containers. However, as its business expanded, the need for computing resources increased, and Momenta faced scaling, availability, and cost challenges. The company found that increasing the number of worker nodes to support business growth delivered diminishing returns on compute performance and limited Momenta’s ability to scale efficiently and cost-effectively. In addition, because Momenta used open source Kubernetes, which had frequent new releases, the company regularly had to take its cluster offline to make updates, interrupting operations. Moreover, dealing with the scalability and availability challenges, as well providing security for node deployment, was putting an increasing operational and maintenance burden on Momenta. To better support its growth around the world, Momenta decided to migrate to Amazon EKS.

Momenta chose Amazon EKS to manage its containers for several reasons. First, Momenta had already been using a wide range of AWS services, including compute, storage, data analytics, and machine learning, to support its operations. Moving to Amazon EKS gave it a compatible solution. Also important, Amazon EKS is a managed service , which allows Momenta to run Kubernetes on AWS without having to maintain its own Kubernetes control plane. Using multiple AWS Availability Zones, EKS can automatically detect and replace unhealthy nodes and provide on-demand, zero-downtime upgrades and the latest security patches. 

Once Momenta had made the decision to migrate, it was able to move quickly from using a self-managed Kubernetes cluster to Amazon EKS. The company conducted its proof-of-concept (POC) verification of Amazon EKS immediately after the service launched in China in March 2020, and, within 2 months, Momenta had completed its migration. One reason the project moved so quickly was that Momenta worked with AWS early in the POC process to gain a deeper understanding of how the Amazon EKS management mechanism worked. AWS provided Momenta with best practices, such as how to integrate Kubernetes management with other services, like AWS Network Load Balancer, AWS Identity and Access Management (IAM), and Amazon Elastic Compute Cloud (Amazon EC2) Auto Scaling groups, and how to migrate related open-source software to the Amazon EKS Kubernetes cluster.

"Through face-to-face trainings and meetings with multiple AWS container experts, we learned that Amazon EKS is easy to use and can fully support Kubernetes in a production environment,” said Dong Weishan, Research and Development Director at Momenta. “Amazon EKS is completely consistent with CNCF (Cloud Native Computing Foundation) Kubernetes, which means we can seamlessly migrate open-source Kubernetes to Amazon EKS. In addition, AWS provides a number of Kubernetes-related open-source tools, including eksctl, AWS CloudFormation, and AWS Cloud Development Kit (AWS CDK), to strengthen the management of Kubernetes clusters.”

Since completing its migration, Momenta has taken advantage of the flexibility that Amazon EKS worker nodes provides. Momenta primarily uses Amazon EC2 Reserved Instances, but when more computing nodes are needed, Amazon EKS automatically engages Amazon EC2 Spot Instances to expand resources to meet business demands. This not only simplifies Kubernetes operation and maintenance for Momenta, but also the combination of Amazon EC2 Spot Instances and Reserved Instances optimizes workload performance while providing significant cost savings to Momenta for compute.

Amazon EKS also offers an automatic upgrade function, which helps provide Momenta with the availability it needs for its business operations. By initiating an upgrade request on the Amazon EKS operation interface, Momenta can use Amazon EKS to automatically detect and replace control plane instances that are in poor health and provide them with version upgrades and patching. As a result, Amazon EKS completes the entire Kubernetes operation upgrade without affecting Momenta’s business continuity.

In addition, Momenta gets the benefit of the open source community focused on Amazon EKS and Kubernetes where users make suggestions and requests to AWS product developers. Ideas generated in the community often become new features of Amazon EKS—so Momenta has the opportunity to contribute to product development and take advantage of user-driven innovation. Moreover, AWS shares its product roadmap with the Amazon EKS community, making it easy for users to understand the product plan and life cycle as well as help ensure the long-term stability of architecture planning for companies like Momenta.

"Although our current infrastructure support team is small, it is sufficient to support production-level Kubernetes clusters by using Amazon EKS. We can also use on-demand computing resources flexibly and further reduce computing costs with Amazon EC2 Spot Instances,” said Dong Weishan.

To learn more about Amazon EKS, visit aws.amazon.com/eks.