Just a Search Box

The first time I saw Google's homepage, I thought it was broken.

One search box. That was it. Every portal at the time was loud. Infoseek, Lycos, Yahoo, goo. News crammed into the fold, banner ads blinking, category links consuming every pixel. Search rankings were arbitrary. Pay enough and you floated to the top. Deciding which site was actually useful was left entirely to you.

Google stormed in with PageRank. Count the inbound links. Weigh their quality. Let the structure of the web itself decide importance, not a human editor. Simple, but revolutionary. That empty homepage with nothing but a search box was a declaration of confidence in the algorithm.

Behind that box, distributed storage was doing the real work. GFS, MapReduce, Bigtable. Google published these as papers, and they descended into the open-source world as Hadoop and HBase. Crawl the entire web. Build the index. Return results in milliseconds. The infrastructure philosophy that made this possible shed its veil in the form of published papers.

BigQuery sits on the same lineage. Query petabytes with SQL. No indexes needed. Full scans by design. In the RDBMS worldview, a full scan is a sin. On Google's distributed foundation, it becomes the optimal answer. Dremel's engine rides on Colossus, thousands of workers reading columnar-compressed data in parallel. The technology forged behind that search box is now available to anyone with a console.

A company that changed the world with one search box now returns petabytes with one SQL query. The brilliant minds once drawn by "Don't be evil" seem busy these days building an empire.