« Windows on Intel Mac's | Main | Frozen in Moscow »
January 13, 2006
Google: where it all begun
The Top 500 Supercomputers website has a list with the most powerful machines in the world according to the popular benchmark suite Linpack. They have machines from US, Japan, Russia or China. They have supercomputers with a few very fast processors and computers with hundred of thousands slower CPUs. Yet, from the top 500 fastest number crunchers, there is an important omission.
Yes, you've guessed, it's Google.
Recent information on the exact number of computers which comprise Google's mega search engine is lacking, but it's not hard to estimate. Back in 2003, Google was reported to have several clusters of as much as 15,000 machines. In 2004, the total number of servers which run Google around the world was already at about 100,000. Ever since, it could have easily tripled and that's not all. At Google, where all servers are made out of commodity parts, hardware often fails. Actually, the whole system is built onto the assumption that hardware fails - as a thing which routinely happens and not as an exception. So new hardware gets added every month and older, slower CPUs get replaced with faster ones, which utilize less power, such as AMD's Dual Core Opterons. Luckily, the robust system redundancy takes care of all the failures which would drive any sysadmin mad in a few days.

IBM's Blue Gene/L at LLNL, the world's fastest supercomputer has 2^17 CPUs. Even if Google's machines do not include custom-built top of the line PowerPC 440 CPUs, the overall performance shouldn't be far behind, given the larger number of processors. So Google, which is currently not included in the TOP 500 fastest machines on earth, could very well be the fastest computing installation ever built!
However, that's not all. The Word Wide Web grows every day and the number of services which Google is offering grows at a similar rate. To keep up with the growth, more and more machines must be added every day - and there is no doubt that they actually are.
The Google computing platform takes care of web searches, GMail and in the very near future, of many other interesting services which we can only dare to guess. As Microsoft have probably already found out while designing the new MSN Search, it's no longer about searching, fast and nice and swell. It's about creating the world's most powerful distributed computing device, with the best price/efficiency one can dream of. A device which can be used for just about anything - with web searching being just a simple (yet incomensurably important) application of it.
And it all begun here.
Posted by Costin Raiu at January 13, 2006 10:06 PM