Google's server layout

[a.titus1]

A buddy of mine mentioned how Google's server layout is somewhat unique in that it uses tens of thousands of low end servers to do the majority of its work as opposed to really high end expensive servers. For some reason I just thought of it a few days ago, and had a quick look around.

The basic idea is they cluster thousands of servers at various datacenters, and they have servers that are constantly failing. But because of the redundency and sheer numbers, they don't take a hit in processing. And the servers aren't top of the line, so it actually runs cheaper than a multithousand dollar server rock.

Anyone know of any more information on this? I thought it was rather interesting the approach they took to this.

 

[Nathan]

It comes from their background. Originally being PHD students, it's how a lot of Universities run server farms for number crunching. The uni can't afford the latest and greatest but generally have a lot of old equipment laying around, so they cluster it all to make one big number cruncher.

They would have used a similar model while doing their PHD and just followed on from that, I assume