Ok, I see what you mean now - I guess we were just looking at different uses for redundancy. I was talking purely from the data's point - if the first server failed then you could always rename the 2nd server and give it the ip of the first server but what you're talking about would certainly require clustering.