The current ICC Champions Trophy 2025 match between India and Pakistan was an exhilarating spectacle, which not solely showcased Virat Kohli’s stellar efficiency but in addition set unprecedented information in digital viewership. Whereas the match was a deal with for followers, it was a technical nightmare for the JioHotstar engineering workforce.
The streaming platform reported a staggering 60.2 crore (602 million) views throughout this high-stakes encounter. With the introduction of the free cell subscription function, the engineering workforce needed to put together for almost 50 million simultaneous streams—a feat no streaming service had tried earlier than.
This required a basic rethinking of JioHotstar’s infrastructure, from API dealing with to community optimisation, to make sure a seamless expertise for thousands and thousands of cricket followers. To sum it up, the workforce did a God-level job.
CDNs are the Key
On the coronary heart of JioHotstar’s dwell streaming structure is a fancy however environment friendly system that ensures customers throughout cell, internet, and linked TVs get a clean expertise. When a viewer requests a dwell stream, the request first passes by content material supply networks (CDNs), which act as an exterior API gateway.
These CDNs are essential not only for distributing content material effectively but in addition for dealing with safety checks and routing visitors intelligently. From there, an inside API gateway, supported by Utility Load Balancers, directs the request to the suitable backend service, which fetches knowledge from both a managed or self-hosted database.
With an anticipated spike in visitors throughout the previous couple of overs of the match, this conventional workflow wasn’t going to be sufficient. One of many largest points was dealing with API calls at scale. Upon analysing visitors patterns, the workforce realised from earlier occasions that not all API requests wanted the identical stage of processing energy.
Some, like dwell rating updates and key match moments, might be simply cached and served with minimal computation, whereas others, like consumer authentication and content material personalisation, required direct database queries.
This led to the creation of a brand new CDN area devoted to cacheable requests, permitting JioHotstar to cut back compute load and considerably enhance response occasions.
The inner API gateway, which serves because the entrance door for all requests, was notably resource-intensive. To mitigate this, JioHotstar deployed high-throughput nodes (over 10 Gbps) and enforced topology unfold constraints, guaranteeing that no single node dealt with too many API requests without delay.
Self-managed Kubernetes to EKS
Whereas optimising visitors dealing with was a serious step, JioHotstar additionally needed to rethink how its cloud-based infrastructure scaled.
Beforehand, the platform relied on self-managed Kubernetes clusters, however these methods have been already nearing their limits. In consequence, JioHotstar migrated to Amazon Elastic Kubernetes Service (EKS), which offloaded the burden of cluster administration to AWS and allowed the workforce to concentrate on optimising workloads.
Nevertheless, migrating to EKS launched new challenges, notably round community throughput. Some of the urgent points was NAT Gateway congestion—a bottleneck that restricted the pace at which knowledge may stream.
In a typical cloud setup, a single NAT Gateway per availability zone (AZ) handles visitors for a number of companies. Nevertheless, with thousands and thousands of customers streaming concurrently, this setup rapidly overloads. To resolve this, the workforce shifted to a subnet-level NAT Gateway configuration, successfully distributing visitors extra evenly throughout the community and eliminating the bottleneck.
Even inside Kubernetes, scaling wasn’t so simple as including extra nodes. Throughout peak load testing, the engineering workforce found that a number of backend companies have been consuming as much as 9 Gbps of bandwidth per node, creating uneven visitors distribution throughout clusters.
Whereas infrastructure optimisations performed an important function in enabling scale, community constraints almost derailed the trouble. Throughout inside load checks, the workforce encountered a crucial IP tackle scarcity in its Kubernetes clusters.
Regardless of configuring personal subnets throughout a number of AZs, JioHotstar discovered that it was unable to scale past 350 nodes—far beneath the 400+ required to help peak visitors. The wrongdoer? Over-provisioned IP tackle allocations.
One of many ultimate hurdles got here from Kubernetes service discovery. Whereas scaling past 1,000 pods, JioHotstar found a tough restrict in Kubernetes’ endpoints API, which tracks community places for companies.
As soon as the restrict was exceeded, Kubernetes truncated endpoint knowledge, creating unpredictable visitors distribution points. Although fashionable EndpointSlices supply an answer, JioHotstar’s API Gateway didn’t help them, forcing the workforce to vertically scale companies to remain beneath the 1,000-pod threshold.
Autoscaling Wasn’t Sufficient
Autoscaling struggles to deal with sudden visitors surges. For main cricket matches, JioHotstar experiences spikes of almost 1 million customers per minute, drastically rising the variety of lively viewers. If a star batsman will get out, visitors can drop by thousands and thousands throughout the similar minute, placing immense pressure on backend companies.
An uncommon problem right here is that when customers hit the again button as an alternative of closing the browser, they’re redirected to the homepage. If the homepage isn’t designed to deal with excessive visitors, it may trigger system failures.
There are further considerations. What if AWS lacks the capability in a particular AZ to provision servers? In such circumstances, autoscaling turns into ineffective. Even step-wise scaling, the place 10 servers are added at a time with a goal of scaling from 100 to 800, could also be too gradual to answer real-time demand.
With 1 million requests per second and 10 terabytes of video bandwidth consumption per second, amounting to 75% of India’s whole web bandwidth, the dimensions of operations is staggering. Notably, even web service suppliers (ISPs) wrestle to ship such huge visitors hundreds.
Streaming platforms continuously encounter these challenges throughout high-profile occasions like IPL finals, royal weddings, or political broadcasts. To arrange for such spikes, JioHotstar conducts intensive load testing utilizing 3,000 machines, every geared up with 36 CPUs and 72GB of RAM, throughout a number of areas.
This rigorous testing course of, often known as ‘tsunami testing’, helps decide the breaking level of every service. The outcomes are then used to plan infrastructure scaling successfully.
The put up India vs Pak: How JioHotstar Pulled Off 60 Crore Reside Stream Views appeared first on Analytics India Journal.
