JioHotstar Details Distributed Engineering for Personalized Ad Requests at Scale

Ad Request Workflow Overview

JioHotstar released an engineering overview detailing its ad request workflow. The platform coordinates distributed services to select, deliver, and measure personalized advertisements during video playback. The architecture focuses on making real-time ad decisions within strict latency requirements while supporting large-scale streaming traffic and maintaining playback reliability.

Ad Selection Process

An ad request begins when a viewer reaches an advertisement opportunity. The request includes contextual information such as content metadata, user context, device information, and available inventory. The platform processes this through multiple backend components that evaluate eligible advertisements, apply targeting rules, and generate the final response for the video player.

Decisioning and Latency

The ad decision workflow involves multiple stages. JioHotstar uses a waterfall tiering approach combined with pacing algorithms like PID and SHALE to select a small number of advertisements for a 30-second ad pod from thousands of candidates. These calculations balance campaign delivery, inventory allocation, and advertiser constraints, completing ad selection and response generation within 100 milliseconds, even during high concurrency events like live sports.

Engineering Challenges and Ad Formats

Pawan Nagar, Engineering Leader, noted that ad serving involves challenges beyond API communication, including managing low latency, handling high traffic, using caching, and ensuring service reliability. The platform supports various advertising formats, including pre-roll and mid-roll video advertisements, as well as display-based formats. After delivery, systems collect signals like impressions and engagement for campaign reporting and measurement.