kjones4 commented: "I have no answers, but I've been kicking around thoughts on hosting too. Have you considered Amazon's S3 + compute service? It looks to be reasonably cost effective and reliable, but I have no experience using it. The ISP business has changed a lot in the last 15 years. It used to be that I wouldn't hesitate to buy a couple of servers, deploy them in a data center, and manage everything myself. Now, it seems to make more sense to be a virtual ISP, where you buy services from people who know better about specific services such as backup, email, fail-over, etc. The problem is knowing if these 3rd party services are in fact more reliable and flexible enough to do the job. My brain hurts with the thought of researching all of this."
Hmm. First, nope, haven't seriously considered Amazon, or LoudCloud, or IBM's "compute as a service thing", or Sun's Grid. I guess I mentally lumped Amazon's things with the other "we have a giant computer, and we'll sell you some of it, and oh by the way, our computer is really big services".
That's what used to be imagined as the future, as far back was the late 1960s, when MULTICS was being written for just that purpose.
Oddly, I think hardware is being outsourced more (outsourced data centers, dedicated hosts, Xen slices/semi-dedicated host, etc). But the *software* is being outsourced less. People hate shared hosting. And those that don't should. The config is bizarre (SSL and plain http on different machines -- I've actually seen that), the systems aren't maintained, there are too many restrictions (can't daemonize apps, have to use a klunky GUI admin to do sysadmin tasks), etc. Shared hosting is ghetto hosting. The jobs of the sysadmin used to be keeping things secure, audited, updated, logs rotated, etc. All of those tasks have either been given up on or automated. Updates are cron job calls to apt-get. Log rotation is from cron and people stopped reading logs a long time ago. I doubt criminals even bother to cover their tracks any more when they penetrate a system, or if they do, it's just to hide themselves from automated break in audit tools. Since the sysadmin is now redundant, mostly fascilities of the operating system, people want to pick their OS. They want to the freedom to pick FreeBSD, Slackware, Ubuntu, CentOS, or whatever for their host. Things like slicehost, where you get your own Xen slice, are profilerating rapidly. There are hundreds of them now.
I don't mean to pretend I wasn't or am not seriously considering shared hosting, but the question of Amazon's and these other "compute cloud" services that came out recently is kind of interesting.
What is their market? A bunch of little guys? They don't do much in the way of support, and the setup is pretty technical.
Big guys? They hate not controlling their own datacenter, or at least racks in a datacenter. Even companies that have a whole bunch of machines at a colo will have their own little datacenter in the office, too. And it's not just for generic compute cycles: there are Windows/x86 machines for domain controllers and all too often, ExchangeServer (ewwwch). Novell servers -- NetWare, I mean. And so on. All of this could be moved to generic compute clusters, but it would require switching products, doing countless "upgrades", and homogonizing things. Hell, IBM must have an ulterior motive of trying to sell their compute cloud in just getting you locked in to AIX by making you switch to AIX software and replacing your database, CRM, etc, etc. Their TV ads are pretty adamant that everything will run there. And every few days, they issue a press release telling you how much power you can save by replacing all of your computers with a big IBM one that's faster than all of them put together. Similar for Sun. As far as I'm concerned, any restriction which keeps the idiot suits at the top from buying and running whatever stupid piece of software they want, is a doomed proposition.