Cheap Hardware + Fault Tolerance = Web Site

The Web is too large to fit on a single machine, so it's no surprise that searching the Web requires the coordination of many machines, too. A single Google query may touch over a thousand machines before the results are returned to the user, all in a fraction of a second.

With all those machines, the opportunities for parallelism and distributed computation are offset by the likelihood of hardware failure. If one machine breaks on average every few years, a pool of a thousand machines will have machines break on a daily basis. A key part of the Google story is that by designing a system to cope with breakage, we can provide not only robustness, but also parallelizability, efficiency, and economies of scale.

Rob Pike is a member of the Systems Lab at Google, Inc. In 1981, while at Bell Labs, he wrote the first bitmap window system for UNIX. He has since written a dozen more. He is a principal designer and implementer of the Plan 9 and Inferno operating systems and co-author with Brian Kernighan of The UNIX Programming Environment and The Practice of Programming.