Fire TV at Amazon Prime Video Modernizes Its Stack Using Amazon ECS with AWS Fargate

Amazon.com Inc. launched Prime Video as a streaming service in 2011. As Prime Video grew in popularity, its teams built more experiences that would support different device types, with one backend presentation service supporting three client applications—iOS, Android, and Fire TV. This backend also supported all 20 unique digital experiences for Fire TV. 

A separate Prime Video team was responsible for scaling this presentation service using Amazon Elastic Compute Cloud (Amazon EC2), which provides secure and resizable compute capacity for virtually any workload. With this model, a bug on one client application could impact the viewing experience across other device types. To overcome this challenge, the Fire TV team at Prime Video adopted a backend-for-front-end architecture model and set up over 4,000 containers using Amazon ECS with AWS Fargate. “Our code can live in a dedicated backend for each device type that we support,” says Sanket Patel, senior software development manager at Prime Video. “We can make changes to the code very quickly and more simply.” 

The Fire TV team at Prime Video also decided to containerize its workloads and take a serverless-first approach—a decision that would help their engineers focus on feature delivery. “Instead of having engineers managing Amazon EC2 instances, we wanted them to work on features that would impact customers and add value to our business,” says Patel. To meet its performance requirements, the team worked alongside the AWS team to increase the soft quota limits for Amazon ECS.

The Fire TV team at Prime Video refactored its application code to a cellular architecture using the AWS Cloud Development Kit (AWS CDK), which gave the team the ability to define its cloud application resources using familiar programming languages. By isolating its presentation service into cells, the Fire TV team at Prime Video reduced the impact of code changes across other Prime Video device types and established clear ownership for its presentation stack. As a result, the Fire TV team at Prime Video freed up its engineers from manual, repetitive work, improving their overall satisfaction and giving them more opportunities to innovate.

The Fire TV team at Prime Video also uses AWS CDK to support personal developer stacks with a self-service strategy so that its engineers can focus on developing features that improve the viewing experience and optimize the performance of their code. “Teams can write tests against experimental production deployments,” says Hani Suleiman, senior principal engineer at Prime Video. “Any engineer can deploy compute resources in their own AWS account and replicate that setup to production.” Empowering its engineers in this way makes it possible for the Fire TV team at Prime Video to verify its code in a production environment within minutes.

To automate the scaling of its infrastructure, the Fire TV team at Prime Video implemented Amazon ECS scheduling tasks and capacity management features. If one of the hosts fails, Amazon ECS with AWS Fargate will automatically replace it within 2–3 minutes. During peak events, such as Thursday Night Football, the team has seen its stack scale to up to 6,000 containers automatically. To further enhance scalability, the Fire TV team at Prime Video set up individual cells for each Fire TV experience. For instance, the team has a dedicated backend for Fire TV devices that support touch interactions and remote directional pads. “Our Fire TV experiences have slightly different data requirements,” says Brent Marieb, senior software engineer at Prime Video. “By having a dedicated cell for every experience, we can specify the number of containers that are required for each cell and reduce the scope of impact of code misconfigurations.”

The Fire TV team at Prime Video integrated Amazon ECS with Amazon CloudWatch, which collects and visualizes near-real-time logs, metrics, and event data in automated dashboards to streamline infrastructure and application maintenance. The team monitors all key performance indicators and metrics using Amazon CloudWatch and can troubleshoot bugs and anomalies faster than before with automated alerts. “We can get metrics at subminute granularity,” says Patel. “If a customer reports an issue to us, we can go to our Amazon CloudWatch dashboards and identify the root cause of the issue.”