Amazon offers dedicated servers on EC2

Maybe we should call this one fog computing?

Over the weekend, Amazon Web Services, the cloud computing subsidiary of the online retailer by the same name, announced something called Dedicated Instances for its Virtual Private Cloud services.

In short, it takes what is supposed to be a multitenant cloud and not only locks access to it through a virtual private network, but also ensures that no one else can run their cloudy apps on a physical server that has your cloudy apps running on it.

In essence, the Dedicated Instance turns Amazon's Elastic Compute Cloud (EC2) and the other elements of the VPC service into a dedicated server hosting plan that has an on-off switch and no long-term commitment. (It's the difference between having full-time girlfriend and a high-class escort service.) The Dedicated Instance presumably doesn't literally guarantee that you have the same exact physical server every time you fire it up after shutting it down.

Amazon AWS is in the cloud computing business to make money, so this Dedicated Instance is by no means free. First of all, the Dedicated Instances are more expensive than regular EC2 instances. A small EC2 on-demand instance running Unix or Linux will run you 8.5 cents per hour and Windows costs 12 cents per hour. But if you want the dedicated instance in a small configuration, you have to pay $10 per hour per region to get the iron dedicated to you and then you have to pay 10.5 cents per hour for the small instance running Unix or Linux and 14.5 cents per hour for Windows.

So there is a two cent premium on the instance plus the 10 bucks per hour. That's a 23.5 per cent premium for equivalent performance, plus the $10 per hour to "own" the system for the hour. (You can see standard EC2 slice pricing here; dedicated instance pricing is there.)

That 23.5 per cent premium carries through on the raw performance for large and extra large instances; for extra large, double extra large, and quadruple extra large instances with high memory capacity; and for medium and extra large high-CPU instances. And so does that $10 per dedicated instance fee per region.

The cluster and cluster/GPU instances sold by Amazon are not listed as dedicated instances on AWS price list, so we assumed that they would be coming in the future when we originally wrote this story. But Amazon says that the cluster and cluster/GPU instances are indeed dedicated by default, as we said you would have expected.

It is not clear if Amazon is allowing multiple EC2 slices to be configured on the same physical server - for instance, if you wanted for some reason to put four small instances on the same physical server and run them in dedicated mode. Amazon may do this by default to minimize the number of physical machines that get locked down by one customer, and then again, it may not. We have an email into Amazon to find out as we go to press. It seems wasteful to lock a user out of any excess capacity in the machine used in a dedicated instance, particularly when they are shelling out $10 per hour to have the machine dedicated to them. But it may be set up that way.

For a quadruple extra large instance, which costs $2.48 per hour running Unix or Linux or $3.08 per hour to run Windows, plus the dedicated instance $10 per hour fee, customers are looking at $109,324 to lock down the instance for a whole year in on-demand mode. That is a phenomenally stupid amount of money to pay for a single server, of course, but remember that the $10 per hour fee only hits you once per region for dedicated instances.

So if you one or a thousand EC2 instances, you only pay the $87,600 for a year one time. And to shave your bill further, you can reserve your instances for a fee that ranges from $280 per year on a small instance to $6,600 for a high-memory, quadruple extra large instance. If you reserve your instances, the price per hour on top of this reservation fee comes way down.

Let's take one example. Say that you want to set up 1,000 dedicated EC2 high-memory instances with quadruple extra large computing capacity running Linux. If you want to do this using standard on-demand EC2 pricing, it's $2 per hour per instance, or $17.52m per year in the US North Virginia data center. Now, you want to say "mine mine mine" and not let anyone else onto the iron running your instances, first reserve the servers for $6,600 a pop, or $6.6m. Then you pay to pay 84 cents per hour per instance to use them running Linux, which is another $7.36m.

Now, toss in the dedicated instance fee, for $10 per hour per region, which is $87,500. Yup, that dedication is noise in the data. And by the way, those 1,000 dedicated, reserved instances only run just a little over $14m, and the dedicated instance represents six-tenths of a percent of the cost of that setup.

The question you have to ask yourself is this: can you buy, run, and support 1,000 servers for less than $14m a year? Many people will say they can, and many companies undoubtedly do. Some companies - particularly startups - simply won't ever want to try.

In addition to announcing the dedicated instances, AWS is also now supporting Microsoft's Windows Server 2008 R2 in a number of Amazon Machine Image formats. You can't yet run it inside of the Virtual Private Cloud setup, says Amazon, but they are working on it. Windows Server 2008 R2 comes in four different images: Core, Base, IIS and SQL Server Express, and SQL Server Standard. Those SQL AMIs include Microsoft's SQL Server 2008 R2 database. Pricing for the Windows instances remains unchanged. You can find out more about the Windows instances here. ®