So you thought Google was a search… advertising … software … information …etc. company? Well, you’re in for a surprise:
“Google is the world’s fourth-largest maker of computer servers, after Dell, Hewlett-Packard and I.B.M.” -says Martin Reynolds, an analyst with the Gartner Group.
Yes, that’s right. Google believes that “its growing cadre of world-class computer scientists can design a network of machines that can store and process more information more efficiently than anyone else. Mr. Reynolds estimated that Google’s computing costs are half those of other large Internet companies and a tenth those of traditional corporate technology users. ” They built hundreds of thousands of servers, typically skipping the box, laying down components on a bed of cork, more recently fastening them together with velcro. Currently they are AMD’s fifth largest customer, and there is speculation that they mey be getting into designing their own chip.
Read the full story at the New York Times.
Your blog requested authorisation for some reason (username and password) I just dismissed it. Thought I’d let you know as this is unsual for a public blog. You might want to double check you http auth settings.
unless this is some kind on new anti comment spam technique..
regards
Al
When RAID emerged there were doubts about it, primarily FUD from the premium storage providers. But as history has shown us, RAID was an exceptional breakthrough for storage reliability. It still runs counter to intelligence in many ways, Centering one’s design on failure stills seems strange.
The concept of arrays of low cost servers/components seems similarly frightening from a design perspective, but in reality, for the kind of uptimes needed for web applications, it must be centered into anyreliable design. Most server manufacturers however are likely to oppose this failure centric design. After all selling a product designed to fail would not appeal to the big margin server marketeers.
It would be great if a manufacturer could start up with a RAID designed cluster. Sell its servers in small clusters with failover designed in, so that components could be easily swapped on failure (accessable components). Of course for this to work each server and it’s components should be “Inexpensive” as in the “I” in RAID, unfortunately most server manafacturers go for Expensive..
regards
Al
Thats odd, I read that article on reddit a few hours prior to seeing this zoliblog post on here using the same image from the first post. Nothing new just rehasing a story to draw attention to a lame blog.
Sarge, thanks for commenting. I make no secret of the fact the the NYT article was published before mine, in fact I am openly using them as source and refer readers to the full article. What is your point?
point is that coming to this page first is a complete waste of time.
Anonymous, are you Sarge? It’s not easy to figure out what your problem is, unless you say it… but I think I’ve got it.
My post got submitted to reddit, and I guess you must have seen it there (?). But I did not know about it, I just wrote my own post, that’s all.