JioHotstar's Distributed Architecture for Personalized Ad Serving at Scale
This article details JioHotstar's distributed engineering approach to serving personalized ad requests within strict latency requirements at streaming scale. It highlights the multi-stage ad decision workflow, balancing campaign delivery, inventory, and advertiser constraints. The system emphasizes low latency, high traffic handling, caching, and reliability across numerous backend services.
Read original on InfoQ ArchitectureIntroduction to Personalized Ad Serving
JioHotstar's ad request workflow is a prime example of a complex, distributed system designed to deliver personalized advertisements in real-time during video playback. The core challenge lies in making swift ad decisions (within 100 milliseconds) while handling massive streaming traffic, ensuring playback reliability, and balancing diverse business requirements.
Multi-Stage Ad Decision Workflow
The ad decision process is not a simple lookup but involves multiple backend components coordinating to evaluate eligible ads, apply targeting rules, and generate a final response. This orchestration is critical for handling various ad formats (pre-roll, mid-roll) and ensuring relevant ads are shown.
- Contextual Information Collection: Upon an ad opportunity, the request gathers content metadata, user context, device information, and available ad inventory details.
- Ad Selection Algorithms: A waterfall tiering approach combined with pacing algorithms like PID and SHALE is used to select a small number of ads for a 30-second ad pod from thousands of candidates. These algorithms help balance campaign delivery and inventory allocation.
- Response Generation: The selected ads are packaged into a response returned to the video player.
Key Engineering Challenges
Pawan Nagar, an Engineering Leader, highlights that beyond API communication, the real backend challenges in ad serving include managing low latency, handling huge traffic, utilizing caching, and maintaining service reliability.
Distributed System Considerations
The architecture requires coordinating services responsible for ad inventory, decisioning, content metadata, tracking, and analytics, all while maintaining low latency. Robust mechanisms for handling failures, retries, and partial service availability are essential to prevent individual component issues from disrupting the user's playback experience. This necessitates close collaboration between advertising, platform engineering, and reliability teams.
While industry standards like OpenRTB define protocols for programmatic advertising, streaming platforms typically build additional internal services for deep personalization, content-aware targeting, and proprietary business logic to differentiate their offerings.