Rental Search Ranking System Design
Explore how to design a rental search ranking system by framing it as a machine learning problem with booking likelihood prediction. Understand the data scale, feature processing, low-latency inference, and methods to scale the system to serve millions of requests per second. Gain insight into practical tradeoffs like feature storage, retraining strategy, and monitoring metrics for production-grade ranking services.
4. Calculation & estimation
Assumptions
- 100 million monthly active users
- On average, users book rental homes 5 times per year. Users see about 30 rentals from the search result before booking.
- There are 5 * 30 * or 15 billion observations/samples per year or 1.25 billion samples per month.
Data size
- Assume there are hundreds of features per sample. For the sake of simplicity, each row takes 500 bytes to store.
- Total data size: 500 * 1.25 *