How Dream11 manages traffic surge
The fantasy sports platform is managing millions of concurrent users while it runs thousands of live experiments to improve user experience.
With close to 110 million users and products such as Dream11, FanCode, DreamX, Dream Sports, Dream Sports (Dream11’s parent company) is India’s leading sports technology company.
During the 2020 IPL, Dream11 managed 60 million requests per minute and witnessed above 8000 contest joins per second.
Handling these gigantic numbers is not an easy feat and according to the company’s CIO, Abhishek Ravi, believes the early adoption of cloud has been one of the most critical parts of the company’s digital strategy.
“This gave us assurance for long-term scale with resilience. Given the elastic traffic pattern that Dream11 witnesses, we have been serving the state-of-the-art user experience with optimised cost,” he said.
Ravi believes the company is very strong in managing traffic at scale. In the latest IPL edition, Dream11 witnessed 5.5 Million users concurrency in one of the matches.
For such cases (traffic surge), the company has deployed two strategies – predictive and reactive.
An automated platform predicts the expected traffic for every match, and the automated scripts take care of baseline IT Infrastructure Deployment dynamically every hour/minute level.
“We ensure that our microservices are horizontally scalable with auto-scaling enabled so that it can scale in case of sudden load or back pressure. We continuously test our load environments to get the right sizing,” said Ravi.
For its Data Stores and stateful systems, scaling is done well in advance according to traffic predictions. To set these stacks to scale automatically, in house platforms have been developed.
Using AI for a better CX
For Dream11, through data analytics, the company has already derived many product enhancement insights such as user personalisation, enhanced gameplay, real-time leaderboard and social features.
“We use data analytics for all our product and technology prioritisation which has helped us in making critical decisions. We run a plethora of experiments, and the data helps us derive the strategy we should roll out for better user experience and personalisation,” maintained Ravi.
Dream Sports is also leveraging deep learning models to analyse consumer behaviour and create a personalised user experience. “Deep Reinforcement Learning is used widely to run 1000’s of experiments at all times on Dream11 platform, to evaluate their effects further and incorporate the learnings near real-time. This, alongside the widespread usage of experimentation, makes every decision consumer-focused,” Ravi added.
AI/ML is used to detect any anomalous behaviour across the system and generate alerts and sometimes take self-corrective actions.
To ensure data security, Dream Sports has deployed both preventive and detective controls in place to monitor its applications and infrastructure.
The company’s network is segregated at multiple levels and monitored real-time. Consumer data is stored in a highly secure and encrypted database with restricted access governed by strict role-based access policy.
“Additionally, we follow the in-depth defence approach where only the necessary services are exposed to the internet and access to them is monitored for any security threat. We also have automation and alerts in place for monitoring and alerting any abnormal activity,” he asserted.