Jeff Bezos (of Amazon) is one of my favorite gurus. The guy has charisma, vision and is very smart. He was, on many occasions ahead of his time and it proved to be very lucrative on all (as far as I know) of them.

Having said that, I must say I’m an avid hardware and hosting freak. I run more than 200 servers in different locations around the world and a lot of those servers I built with my bare hands (yep, I risk the static electricity stuff; you don’t look quite so cool putting in dimms with a little bracelet on; never went wrong so far. knock-on-wood).

I run mostly Debian on my systems (Etch or Sarge) and it usually runs fine. We don’t have many hardware breakdowns (buy expensive hardware; it really is better) and be nice to your systems; don’t overload them, use monitoring etc. Problem is, small sites sometimes grow big and then the scaling thing starts. Although we don’t sell cheap hosting accounts, we don’t have resources to buy SANs, solid hardware load balancers etc. We do everything with open source software and it works, mostly.

It works until the real scaling starts. Databases grow, diskspace get’s eaten away by media files, iostat tells me 99-100% all the time now on all nodes etc and those kinds of things. This doesn’t happen with ‘normal’ sites, but with some sites I host, this is more rule than exception.

When it comes to big media files, we wrote our own software solution. It works like Amazon S3, kind of and it is quite robust handling the millions of media files across nodes, serving them up fast. However for databases we, like so many, did not find a solution yet. In our brains we have one, but that is not working (yet).

So I decided to check out what the current ‘cloud hosters’ have to offer. What my dream of hosting always has been and still is; get an account for a site, attach some domains (DNS managed by the package you get), click run and that’s it. 100% uptime, whatever size it grows too, how many db connections, how big the db etc will just scale. It’ll scale, you pay. Never any software tweaking/fixing, no more hassle with moving stuff to different servers, no more partitions, no more… dream…

Anyway; I actually like tweaking and optimizing, but a bit less hands on in the server room would be nice and my hosting dream is based on that. Who delivers that? No one indeed.

I started, as I like hardware etc, requesting offers for this ‘kind’ of solution. A few hosters were very accommodating, notably Rackspace; they wrote down a whole plan for arranging this kind of thing. Ofcourse I gave them some input to the scaling parts and they were very helpful in working it out for me. The end result was upwards of $50.000 / month unfortunately. Not that expensive in fact, but too much compared to the income from the services I am delivering.

Dedicated servers, and, in fact, most dedicated server companies didn’t seem to really mesh with the idea of scaling. They just want to move servers and don’t care at all what happens after that.

Because my services deliver more money to me when they are used more, I would like something that scales with my needs, meaning, in my mind, I need ‘cloud hosting’.

The service that looks to be closest to my hosting dream is Mosso or Google AppEngine, after that GoGrid and after Amazon EC2. This solely based on marketing materials. Which are, ofcourse, really not truthful.

As Google doesn’t support enough ‘features’ yet for me to work with (we have our own, enormous, framework and really need to run that; Google cannot do that, yet), I skipped them and went on to Mosso.

Mosso seemed very nice. Install-and-forget. Just get their site package for $100, point your domain to their DNS servers, use their hosting panel to get all up and running and there you go. Great fun. It is. If you don’t need anything else than they installed on your image. As you don’t get root access, what they put on their for you in terms of OS software like Apache, Ruby, Python, Perl, PHP and MySQL is it. You cannot touch any of it. Either your stuff is there or not. The presales guy told me, via the phone and chat (I have records!) that all popular stuff, including what I needed, was on their systems. Unfortunately, after buying, it turned out there was no way to convert videos; no ffmpeg or anything like it. Making it usable only if I use one of my other systems as conversion webservice. Although that is still an option for me, this kind of took the wind out of my sails.

Naive as I am, I always think someone did something magical to MySQL to make it scale without intensively rewriting/fixing my software.  Ofcourse this also turned out not to be the case. Their MySQL severs are tweaked, but they have nothing special; they can (meaning the will) break.

I still like the Mosso idea and philosophy and I would advice people to try it out when looking to run their standard site on a cloud, but for me it was not really happening at the moment.

GoGrid seems to be Amazon, but more userfriendly and with more features. Their pricing model is based on memory use and outbound traffic; diskspace and inbound traffic are ‘free’. Great for backup, right? Not really ofcourse, these guys at ServePath are not insane. Memory usage is linked to the diskspace you get assigned. So when you get a 512mb server, you end up with 25 gb which you cannot (!) enlarge without creating a new server. Nor can you snap one server to another; hell you cannot take snapshots at all. They provide no backup options or space (just start another server for that; meaning, spend a lot of money for a backup). For web traffic load balancing/failover they provide a free load balancer (nice), but for MySQL they couldn’t give me any options except ‘get a custom solution at ServePath’.

All in all; GoGrid is an operation that can turn out nice, but they are miles away from providing anything more than a, very basic, VPS management system. Throwing HyperVM + XEN/OpenVZ on a few machines gives you *a lot* more for a lot less.

I ended up at Amazon. I am a programmer and a Linux admin, but I really don’t like weird proprietary tools to manage systems. Especially when they are written in *cough* *cough* Java. If I do use those kind of tools I want to see a nice shiny interface and do a bit of clicking.

The setup seemed annoying to me at first, but as usual, I put everything down in scripts the first time I do it and, if I am in a good mood, and I was, I put a web interface on top of them. After a few days of fooling around with EC2 (and such) I had a webinterface to:

• Manage machine instances (create/destroy)
• Set/Reset root  (and other) passwords
• Install tools on the instances automatically
• Create Elastic Block stores on instances and mount them
• Take/Restore snapshots
• Create/Destory AMI’s

I know there are open source tools to do this, but they really weren’t that well suited to what I set out to do and I needed to learn everything fast which I did this way.

Things To Do are automated snapshotting for fail over, load balancer/fail over installation, but I’m already quite happy with this.

Personally I am convinced this kind of thing is the future. It is not the cheapest solution, but it is really scalable, easy to use, backed by a big company and created for building large scale applications and infrastructure on. Although they were down a few time (they were/are in beta stage with some of their products) the idea is very solid and tempting to use. I will give it a try with one of my main services and let you know.