Reducing Developer Overhead by 50% by Migrating to Amazon EKS with Mobileye

Mobileye leads the mobility revolution with its autonomous-driving and driver-assist technologies, harnessing world-renowned expertise in computer vision, artificial intelligence, mapping, and data analysis. As of October 2023, more than 150 million vehicles worldwide have been built with Mobileye technology inside. Mobileye’s REM team focuses on creating high-definition road maps that autonomous-driving vehicles use for navigation.

The REM team built its infrastructure using a variety of open-source tools, such as Apache Spark. However, its previous software stack provided little flexibility or configurability. “We wanted the ability to make changes on the fly without redeploying or reconfiguring compute resources,” says Ben-Tzion Lifshitz, software developer at Mobileye. “We couldn’t configure different Apache Spark drivers or diverse compute sizes in a single cluster.” As a result, changes that engineers made to the configuration of one application could impact other applications in production. Mobileye’s engineering teams will benefit from each application running in an isolated environment. “It will dramatically reduce our debugging times,” says Ofer Eliassaf, senior director of REM at Mobileye. “The approach helps us avoid situations where one Spark application interferes with another, freeing up our engineers to solve complex problems.”

Mobileye decided to migrate its workloads to Apache Spark on Kubernetes using Amazon EKS. “Using Amazon EKS, we removed the additional burden from our DevOps teams,” says Eliassaf. “Kubernetes is very complex software, and using Amazon EKS makes it simpler to install and work with Kubernetes than I had imagined possible.”

In February 2023, Mobileye began migrating to Amazon EKS, testing the performance of small-scale workloads. For its machine learning and data workloads, the REM team shortened its development time by using Data on EKS (DoEKS), an open-source project aimed at streamlining and accelerating the process of building, deploying, and scaling data workloads on Amazon EKS. Mobileye also engaged the AWS team for technical support with scaling. “With the support of the AWS team, we adapted configurations that worked best for our existing environment,” says Lifshitz.

Mobileye also used DoEKS Blueprints, which incorporate guidance and tools to support the challenges and requirements of data-intensive workloads on Kubernetes. The REM team used DoEKS Blueprints to configure Kubernetes to deploy Apache Spark solution to run with the Spark operator.

The team configured Amazon EKS to run alongside Karpenter. By using Karpenter, Mobileye has unlocked automatic scaling capabilities and can diversify its instance types in Amazon Elastic Compute Cloud (Amazon EC2), which provides secure and resizable compute capacity for virtually any workload. For example, the company runs the majority of its workloads using Amazon EC2 Spot Instances, which businesses can use to take advantage of unused Amazon EC2 capacity in the AWS Cloud at up to a 90 percent discount compared with Amazon EC2 On-Demand prices. Using Amazon EKS, Mobileye scaled to run more than 400,000 virtual CPUs concurrently on thousands of Amazon EC2 instances across two clusters. “We can automatically scale much better using more diverse instances,” says Lifshitz. “We can also run less critical jobs on Spot Instances and diminish instances in 10 minutes after we’re finished using them.”

With greater flexibility to configure containers, the REM team can also boost productivity. “Using Amazon EKS, we can change the way our developers work,” says Eliassaf. “We can provide them with a better experience, which will improve their performance and our time to market. They can more easily use open-source tools and AWS services.”