What happens when the host box dies? If you build in redundancy, the costs start spiralling quite quickly...

The chances of that happening are low anyway. I believe someone reported their server was running smooth for 3 years with only a single _minor_ downtime. Also, I doubt that an app that runs on a VPS is as important as one that requires a whole dedicated server. But yes, if a box happens to fail, how slim the chance may be, then your site is offline, just like Amazon EC2 that was offline for 3 days iirc even though they have multiple availability zones per region.

Also, running 3 dedicated servers for redundancy still isn't expensive when you compare it to traditional VPS/EC2. However, you do have to do everything yourself of course. Tbh I feel that people always tend to worry too much about "downtime" or "failing boxes" while the reality is that most people don't ever experience huge downtimes. If the app is really _that_ important then you're likely making some good money to justify paying for more expensive solutions and there wouldn't be a reason not to do that.

Wouldn't the risk of that be exactly the same as the risk of a host box dying at for any VPS product anywhere? Specifically I'm sure Hetzner is running their VPS product on the same boxes, since one of their methods is to have a very standardized, custom designed box (like google does).

The good VPS providers have fail over hosts so that they can move your guest if the primary host is experiencing issues (e.g. RAID degradation). Linode have this for example (I know because one of my 512MB guests got moved recently).

Edit: Leaseweb (who are on a VMWare stack) also offer something similar via Vmotion

I'm no expert, but how many VPSs could you run nicely with just a typical CPU in there?

As many as you like, if you don't mind overcommitting the cores. CPUs overcommit quite well - it's when you start overcommitting RAM that you get into trouble.

But surely you'll have a hell of a lot less CPU power than, say, 128 of Linode's 512mb virtual servers, meaning they would have much worse performance if fully used?

That would depend on the type of CPU Linode is using. What the CPU's frequency is, the amount of cores, whether it has or does not have hyper-threading technology.

The hyper-threading technology is a pretty cheap way to add more virtual cores to your VPS's which is great. One thing is for sure and that is that Linode provides you 4 virtual cores and not 12 per VPS. But, that of course doesn't mean their CPU is better or worse. But, if you can run 6 cores at 3.8ghz, or 12 cores at 1.9ghz then I wouldn't worry too much about not having enough CPU power for your virtual private servers.

If any CPU-overcommitted VM system has guests which all want 100% of the CPU, they'll all suffer. No two ways about that. I don't know what hardware Linode put their 512MB instances on, and I don't know their overcommit ratio. 128 is more guests than I'd put on a 64GB host, though.

If your typical load is actually at 100% CPU for the majority of the time, I'll go out on a limb and suggest that VPS might not be the best bet anyway.

How hard is Xen to configure and maintain? I like VPSs because they're mostly hassle free, and all my projects are separate and isolated. It seems like running all my projects on one box would be a massive headache!