Airtel Delivers a Smooth Streaming Experience for Wynk Studio by Adopting Karpenter on Amazon EKS

Independent artists are stepping into the spotlight thanks to the rise of digital streaming platforms. In 2020, indie artists earned a record high of $1.2 billion. In India, independently funded and promoted artists now share top 10 spots on streaming platforms with their corporate-backed counterparts. On the flip side, indie music yields an added source of revenue for over-the-top (OTT) entertainment platforms, making the streaming game a win-win for artists and companies alike.

Bharti Airtel Limited and its subsidiary Airtel Digital Limited (Airtel), a leading OTT platform in India, launched Wynk Studio in June 2022 as a place for indie artists and podcasters to digitally publish and promote their content. Wynk Studio currently offers 13.5 billion minutes of content delivered to 90 million active subscribers on Wynk Music—India’s largest aggregated premium content marketplace.

Airtel chose to build its backend infrastructure on Amazon Web Services (AWS) for the platform’s range of modern technologies and reliability. It’s using Amazon Elastic Kubernetes Service (Amazon EKS) to orchestrate its microservices-based environment and Amazon CloudFront to stream 1 TB of data per minute through the Wynk Music platform. The OTT provider relies on AWS Enterprise Support for technological guidance as it continues to innovate new products such as Wynk Studio.

Airtel began experiencing highly variable workloads for its internal applications on Amazon EKS, with traffic multiplying two to four times in a matter of seconds. Costs were rapidly increasing and the open-source Kubernetes Cluster Autoscaler Airtel adopted was taking approximately 2 minutes to bring up new nodes when workloads spiked. This led to sporadic delays during content playback on the Wynk platform.

Furthermore, deployment had become overly complex, as Airtel engineers needed to manage many node groups with separate configurations for diverse application requirements. The business consulted with AWS to optimize costs while maintaining, or even improving, availability and to simplify the management and governance of Amazon EKS nodes.

To reduce costs while improving elasticity, Airtel increased its adoption of Amazon EC2 Spot Instances on production and non-production Amazon EKS workloads with Karpenter. The business also implemented AWS Node Termination Handler, an event-driven service, to quickly respond to interruptions when Spot Instances become unavailable.

“We wanted to use more Spot Instances but were initially concerned about how this would impact application availability. With AWS Node Termination Handler, we’ve made our production and non-production environments resilient from Spot instances interruption and termination that can lead to application downtime,” says Sagar Arora, senior principal engineer, Devops at Airtel Digital Limited. The company reports savings of 47 percent on Amazon EC2 costs since increasing Spot coverage and adopting AWS Node Termination Handler. Keeping costs low—both internally and for subscribers—is important to Airtel’s business strategy to monetize its OTT platform.

To further reduce costs and reduce the complexity of managing multiple node groups from the use of Spot and GPU instances, plus multiple CPU architectures, Airtel implemented Karpenter as an open-source autoscaler. Karpenter is an open-source node provisioning solution built for Kubernetes. The solution selects optimal instances for provisioning and terminates instances that are not in use to reduce waste.

Airtel currently uses Karpenter with 90 percent of its Amazon EKS nodes in production on seven Amazon EKS clusters. This has lowered the time required to scale up from 2 minutes to 40 seconds. Scaling down, which used to take 1 minute, now takes just 20 seconds. Arora says, “The simplicity in which instances are assigned to nodes within the cluster, and how easily you can discard nodes to scale down and save costs, are the main advantages of Karpenter.”

Before adopting Karpenter, Airtel’s engineers were managing four node groups with multiple instance types on average. It often took one to two days to upgrade node groups. Now, engineers have just one node group to manage with one Amazon EC2 On-Demand Instance, and upgrades are performed automatically in minutes rather than days. The time engineers formerly spent managing multiple node groups is now spent refining Airtel’s continuous integration and continuous delivery (CI/CD) pipeline.

Arora says, “Karpenter has reduced the time required to orchestrate and deploy new instances according to workload requirements. We no longer have to create and manage separate node groups. Instead, we can create one default provisional configuration that Karpenter uses to cater to the demands of each workload.”

Reliability is the highest priority for the Wynk platform; avoiding downtime is crucial to retain and attract customers. Karpenter helps Airtel deliver a highly available application, with an inherent fallback mechanism that switches to On-Demand Instances on the rare occasion when Spot Instances are unavailable.

Arora explains, “The combination of AWS Node Termination Handler and Karpenter has helped us resolve service interruptions. Karpenter picks the best instances according to pending workloads, choosing instances with the least probability of interruption at the lowest price. This ensures our application is always up and running.”

As an early adopter of Karpenter, Airtel worked closely with AWS on the proof of concept (PoC) to test the solution before implementing it in production. Decisions around configuration, node termination, and Spot interruptions were resolved in a matter of weeks, and the PoC was completed in three months. Airtel’s AWS technical account manager (TAM) and the AWS Enterprise Support team maintained close contact throughout the adoption and cost optimization journey, resolving queries and proactively finding the right solution that worked for Airtel’s specific use cases.

“The AWS Enterprise Support team took our feedback and quickly resolved issues in testing. The solution is now very mature and stable, and we haven’t experienced any issues in production,” says Arora.

More recently, Airtel has focused on bolstering security by building and deploying Amazon Machine Images (AMIs) with necessary security patches. By updating Karpenter provisioner configurations, Amazon EKS worker nodes can be upgraded with these custom AMIs. Karpenter will then automatically pick up new AMIs when nodes are refreshed or recycled.

Airtel is also experimenting with a new pipeline of product launches on AWS, for which it’s inclined to implement Karpenter on Amazon EKS from the start. Arora summarizes, “The Enterprise support we get from AWS helps us move forward with new technologies that promote innovation in our business.”

To learn more, visit aws.amazon.com/solutions/app-development.