Why Googling Instead of Yahooing? - Homebrew Habits
(Page 2 of 4 )
With $9 billion in cash and plans to spend more than $1.5 billion this year on operations centers and technology, Google can afford the best hardware around. But the company hasn’t strayed that far from its roots as a graduate-school project growing in a garage. To this day, most of Google’s servers are custom-built from relatively low-cost hardware.
It was originally a cost-saving move, and even now it saves Google quite a bit of money to build its own. The servers put together by the search engine are designed to use less electricity than traditional machines; the money can really add up when scaled to hundreds of thousands of servers. Google had to cut corners somewhere, though, and, at least for the initial network, it was in reliability. Google’s entire network, in fact, was designed by Larry Page based on the assumption that parts would fail regularly. Hence, Google’s computer racks really are held together with Velcro. Urs Holzle, Google’s senior vice president for operations, has observed that “Nobody builds servers as unreliably as we do.”
What reliability Google’s network may lack in hardware, it makes up for in software. The company’s home-brewed software tools make the most of its distributed network. For example, some of them make it easier to manage parallel processing. Stephen E. Arnold, author of a book about Google’s technology (The Google Legacy), notes that one of the problems with parallel processing is simply getting it to work correctly. “If you talk to guys who work in massively parallel computing operations, as much as 30 percent of their coding time is spent trying to figure out how to get the thing to run.” Google, on the other hand, “has figured out how they can reduce a lot of the hassle and work of creating parallel applications.”
Google has come up with other ways to make the most of its hardware, and accommodate its relative unreliability. For instance, software called the Google File System copies data to several places. This means when one server fails, it doesn’t need to panic about losing the data. It also means that Google doesn’t have to make regular data backups in the same way as many other companies. Now that’s taking lemons and making lemonade!
Google can’t use inexpensive, relatively unreliable servers forever, though. It might work for search, but with email (Gmail) and now ecommerce (Google Checkout) going through the machines, users of its services will get justifiably angry if any of their precious data is lost. Even there, Google is looking to keep costs down. It is one of Advanced Micro Devices’ largest customers, and reportedly favors the power-sparing Opteron chip.
Google is even doing business with Sun. Sun’s systems are expensive but reliable, and the firm recently created a chip that is particularly efficient in its use of electricity. That could add up to a real monetary savings for Google when spread across enough machines.
Next: Google an Also Ran? >>
More Search Engine News Articles
More By Terri Wells
/ 4 
