A day actually is a catastrophe for pretty much anything in the IT world, and that's still about 99.6% uptime. The benchmark a lot of major companies shoot for is 99.999% uptime, which tranalates to about five minutes a year of combined downtime between unexpected outages and "we're rebooting the server, back in a sec" sorts of things.
On the other hand, a friend of mine runs a small email/webhosting business on the side - couple hundred regular customers over several years, that sort of thing - that grew out of a hobby of his when he was younger. It's still an on-the-side thing as he has a fulltime regular job in a different field. He has, oh, five hours or so of downtime annually and doesn't push himself hard to keep it that way, so it's absolutely doable. (Of course, his server's in Brooklyn so he's rather over his annual quota, but is also fully aware that nobody in their right mind's going to hold that against him.)
In practice, reliability higher than that is getting into bragging-rights territory more than anything. There's a few businesses and industries that need that kind of wacky reliability, but not many.
A major site like Facebook or Google or Steam being knocked offline for days is the sort of IT disaster that would require enough preceding things going wrong first that it actually probably would justifiably damage the reputation of whoever it happened to, since it'd be a sign of a pretty significant organizational failure rather than just a single machine going down for awhile. That's why even minor outages with companies that big tend to hit the news.
For some context, DU shoots for around 97% downtime from what I can tell, with most of that 3% being the nightly downtime for backups and the like. That's just fine for a site like this, of course!
= new reply since forum marked as read