Google's First Server Was in a Lego Case
Zusammenfassung
In 1996, Larry Page and Sergey Brin built their first web crawler’s storage system using ten 4 GB hard drives — in a case assembled from Lego bricks. The Lego housing allowed cheap, modular expansion: adding new drives required only adding more bricks. The server crawled the web for the BackRub project, which became Google. The original Lego server is now in the Stanford Computer History collection. Every major data center Google has built since — from the first colocated racks to custom-designed facilities in Finland and Nevada — traces its lineage to a system two graduate students built from children’s toy bricks because they needed cheap, expandable storage.
The Graduate School Infrastructure
Larry Page and Sergey Brin were Stanford PhD students in 1996, running computationally intensive experiments on the university’s research computing infrastructure. Page’s project — analyzing the web’s link structure — required downloading significant portions of the web, which required significant storage.
Stanford’s computer science department provided server room space and network connectivity. The students provided the hardware, purchased with research funding. In 1996, a 4 GB hard drive cost approximately $350. Ten drives — 40 GB total storage — represented a significant investment and required a housing that could be expanded as the project grew.
Commercial server cases were expensive and fixed in size. Page and Brin assembled a custom housing from Lego Technic components: modular, cheap, and easily extended by adding more sections. The result was functional, distinctive, and entirely appropriate for a university research project.
The Transition to Real Infrastructure
By 1998, BackRub had consumed enough of Stanford’s network bandwidth that the university was actively encouraging Page and Brin to either commercialize the project or shut it down. They incorporated Google in September 1998, initially operating from a garage in Menlo Park, then moving to an office in Palo Alto.
The first commercial hardware was purchased with the $100,000 check from Andy Bechtolsheim — standard rack-mounted servers, not Lego. By 2000, Google was running thousands of commodity servers in colocated data centers; by 2005, it was designing its own server hardware for maximum efficiency.
Google’s Data Center Evolution
Google’s approach to hardware infrastructure was shaped directly by the graduate school ethos: commodity components, horizontal scaling, and software-managed redundancy rather than expensive high-reliability hardware. Rather than buying enterprise servers from Sun or HP, Google designed its own motherboards, power supplies, and cooling systems — eliminating every unnecessary component.
This approach — shared with Amazon Web Services and later with every major cloud provider — is the standard model of modern data center operation. The insight that you can build reliable infrastructure from unreliable components through redundancy and software, rather than buying expensive reliable components, traces to the culture of building things from whatever was available and cheap. The infrastructure that resulted is described in The Cloud Wars and the AWS story.