New York Times has an article about Google's innovations in hardware and software that deals with a large number of computers. Google uses cheap computers, that have a high rate of disk failure, so they had to use data redundancy and make multiple copies of the data. They had to build a reliable software to overcome the unreliability of their hardware.
"Google is as much about infrastructure as it is about the search engine," said Martin Reynolds, an analyst from Gartner Group. "They are building an enormous computing resource on a scale that is almost unimaginable." Google is the world's fourth-largest maker of computer servers, after Dell, Hewlett-Packard and I.B.M.
"We don't think our competitors can deploy systems cheaper, faster or at scale," said Alan Eustace, Google's vice president for research and systems engineering. "That will give us a two-, three-, five-year lead."
Google has created some powerful tools for their servers:
* MapReduce, that divides a problem to be handled by thousands of processors simultaneously
* Google File System, a scalable distributed file system for large distributed data-intensive applications, that keeps back-up copies of data in several places. Google File System stores data efficiently.
* Google Work Queue, a job scheduling tool for servers.
Google uses servers with Opteron chip from AMD as they are more power-efficient, but it wants more. "Google's next step is to build high-performance silicon," said Mark Stahlman, an independent technology analyst.