Speed Search: Making Expedia Flights Faster

The motivation was to address the peak search traffic and latency overhead on the Flight Search page, which impacted user experience and bookings. Improving performance was crucial for user retention.

The custom metrics include page usable time, which tracks the time taken for a component to render, and non-supply overhead, which measures the total time minus supply overhead time. Common metrics like First Contentful Paint and Cumulative Layout Shift were also monitored.

Prefetching involves fetching JS bundles beforehand during browser idle time on the homepage, which saves time when the user lands on the Flight Search page. This initiative helped improve CDN path retrieval speed by about 100ms.

Preemptive search predicts the Flight Search response using user inputs like origin and destination on the homepage. The response is cached for faster retrieval when the user lands on the Flight Search page, improving performance by up to 50%.

Microqueries break down bulky queries into smaller subqueries that run in parallel. This allows users to see chunks of search offers immediately, improving performance by about 20%.

The initiative involved analyzing the waterfall diagram to ensure queries were executed in parallel rather than sequentially. This change led to an 8% performance improvement on the Flight Search page.

The Micro-Front-end Architecture breaks page-level components into shareable packages, allowing for efficient monitoring and maintenance. It also helps in reducing bundle size and improves flexibility.

The PERP initiatives led to a 52% improvement in page usable time and a 40% improvement in non-supply head time for the top 90th percentile.