Engineering lessons for traffic spikes: the case of 32M viewers

The 2023 Indian Premier League (IPL) final brought together 32 million concurrent viewers, a record that put JioHotstar's (then JioCinema) streaming infrastructure to the test. Their engineering team, led by Ashutosh Agrawal, shared the keys to ensuring an uninterrupted broadcast at P99 CONF. Although the context is extreme, many of their lessons are applicable to any company facing demand spikes, such as an SME during peak season or an IT team facing a launch.

Preparing for the unthinkable: double capacity

The JioHotstar team did not settle for estimating expected traffic. For the 2023 final, they prepared for between 50 and 75 million viewers, roughly double the capacity they anticipated needing. As Agrawal explained, it is "a scary number that frightens the whole team," but that extra margin allows absorbing spikes without requests piling up. For an SME, this translates into sizing servers and cloud services with headroom, especially before events like Black Friday or the end of a quarter. It is not about always over-provisioning, but about identifying critical moments and acting accordingly.

The golden path: video delivery cannot fail

Agrawal made it clear: "You can sacrifice any other system or component, but everything in the video delivery line must be a golden path; it cannot fail." For an SME, the equivalent would be the checkout process in an online store or authentication in an essential service. It is essential to map all components of that critical path, from authentication to the CDN, and ensure that each link is prepared for the spike. In JioHotstar's case, they detected that the biggest bottleneck was not in the final, but on the first day, when millions of new or inactive users tried to log in within minutes. Therefore, they proactively scaled systems such as SMS and OTP authentication, and established agreements with providers that had to meet strict latency SLAs.

200-point checklist and exhaustive testing

Before the event, the team conducted an audit with a 200-point checklist covering everything from the database version to timeouts between services. Many of those points arose from past incidents, such as a service that had not correctly configured keep-alives and caused the load balancer to crash. For each possible failure, they documented a mitigation strategy. For example, if all SMS providers failed, the system had to allow access somehow, even temporarily without authentication. For an SME, you don't need to reach 200 points, but it is advisable to review critical components and have contingency plans for external providers.

Testing was also key. They conducted vertical tests of each system and provider, and then multi-system load tests to see how they behaved together. As Agrawal noted, "a service can scale perfectly in isolation but reveal its true face once you add dependencies." In our day-to-day, this translates into not limiting yourself to testing components separately, but simulating real scenarios with all systems interconnected.

What it means for your SME

You don't need streaming to millions of people to benefit from these lessons. Any company that has traffic spikes (marketing campaigns, registration openings, product launches) can apply the same principles:

• Size with margin: calculate capacity for at least double the expected load at critical moments.
• Identify your golden path: define which processes cannot fail under any circumstances and reinforce their monitoring and redundancy.
• Review external dependencies: ensure that your providers (payments, SMS, email) meet SLAs and have alternatives.
• Document contingency plans: for each single point of failure, define what you will do if it fails.
• Test together: don't limit yourself to unit tests; simulate real load with all systems integrated.

At ForgeNEX we help SMEs prepare their infrastructure for these scenarios. If you want to delve deeper into how to manage traffic spikes, you can consult our analysis on Cloudflare and application control or review the latest security threats that also affect availability.

The conclusion is clear: meticulous preparation and the mindset of anticipating failures make the difference between a service that holds up and one that goes down. Even if your company doesn't have 32 million users, adopting these practices will allow you to face with guarantees those moments when your entire business is at stake in a few minutes.

Source: The New Stack. Analysis and adaptation: ForgeNEX.