One of our customers asked me why we make use our own metal, rather than using a Cloud service like that provided by Amazon.

I replied that it was a matter of cost - an EC2 instance with sufficient RAM and block storage would cost several times more than renting/leasing/buying a dedicated server.

If it is too expensive to use for hosting, how is it not too expensive to use for ANYTHING that isn't on a massive scale?

I can see the fault-tolarent advantages the technology offers, but I can't see the advantage over having metal-based fault tolarance, or simply having N+1 servers and running vmware or similar, which again would be cheaper.

In fact I can see how using AWS would be less reliable than your own metal - you can't tell if there's a potential fault on the physical server your instance is running on and might go BOOM any moment, potentially corrupting a critical database or file-write in the process. Sure, your instance would restart within moments elsewhere, but if something got corrupted when it went down, it might not work.

I'd be interested in comments on this, or maybe a pointer to a forum elswhere where it might have been discussed in the past.

Very true, amazon does not make sense for any serious level of application hosting. It very quickly will become more expensive. Example case, when we were testing a very large commerce system the benchmark alone would cause the system to scale from 3 to 40 systems immediately. Under real world conditions that 3 host system would probably always run that high.

Secondly Amazons business model is about cramming as many people into the system as possible. Through the magic of "cloud" technology we are apparently able to defeat thermodynamics and get energy from nothing. Dare I say, we have solved the energy crisis!

Reality here is that the environment is exactly what we all think it is. Bare metal systems, distributed storage, virtualization, and some provisioning automation. Basically the exact same thing you'd be using anyway. Their advantage is they have a lot of it, so they have more redundancy (in theory), more access to bandwidth (in theory). What this means is you pay for over-subscribed systems. (We tested this!) They under-perform and for amazon, thats a good thing. You have to buy more of them, and thus they make more.

So why does it get used? The cloudy redundant magic is techie babble to most people. They probably dont need more than what amazon offers: a system to do their thing, perhaps with more access than they'd get in a general hosting environment, and through marketing that "Cloud" == "what they need".

Thanks Scott.

OK, Devil's advocate here: There's one area where AWS seems to excel: S3. As a target to backup to, an S3 bucket is the fastest and cheapest option I have encountered, especially as it is "pay as you use" and getting cheaper ever 6 to 12 months as well.