Google Search - the case study

The whole nine-board set, once you already know the story each piece is telling.

Boards

Before you start. You already know the feeling of typing a couple of words into a search box and getting a good answer back before you’ve finished sitting down. This lesson takes that one second apart and shows you the two separate machines working underneath it, one running all the time in the background and one that only switches on the moment you hit enter.

The idea in 30 seconds

A search engine this size never searches live: it reads, sorts and scores the whole web ahead of time, so your query only has to look something up.
Flipping “what’s on this page” into “which pages hold this word” turns a search into a look-up and a merge, instead of a read of everything.
Splitting that lookup table into thousands of pieces and asking them all at once means your wait is set by whichever piece is slowest, not by the average.
Asking the same question of two copies at once, and keeping whichever answers first, turns a rare slow piece back into a rare event instead of a near-certainty.
The page that wins isn’t the one that repeats your words the most, it’s the one that other trustworthy pages vote for, combined with hundreds of smaller checks so no single trick can win by itself.
Keeping that lookup table fresh without ever fully closing it needed storage and processing tools that didn’t exist yet, which is why so much of today’s large-scale software still traces back to this one problem.

Read the story · Skip to the numbers