Google On How It Maintains Search Reliability And Speed

Google is out with a blog post today detailing its Search infrastructure, and how it keeps in mind speed and reliability when developing new features.

Reliability, or being “available when you need it,” is Google’s top priority for Search. This includes “unforeseen surges” in lookups from global sports moments and other cultural events, as well as “critical searches related to natural disasters.”

Search data scientists continually evaluate subtle signals, like users refreshing a page, to identify cases where Search is not meeting people’s expectations. Engineers then use these signals to identify weaknesses in the system and build mitigations to prevent outages.

Google’s Search server infrastructure is “built to process billions of searches every day… regardless of the capability of your network or device.” That translates to an “average” user having to complete around 150,000 searches before “encountering a failure due to an error in our Search infrastructure.”

That means if you searched 10 times a day, it would likely take you more than forty years before encountering a server-side error.

To maintain speed, Google explains how any “increase in latency (from a new feature or change to Search) must be offset by making some other part of Search faster.” This optimization takes the form of “phasing out slower code and lesser-used features.”

The company specifically measures the time — in milliseconds — “between entering a search and seeing results.” Given the number of users, “trimming time off of individual queries adds up to major time savings for people using Search.” In the past two years, this has resulted in saving “users over 1 million hours every single day.”

In introducing AI Overviews, which are not yet faster than traditional results, Google says it has “saved users another half-million hours daily.”