This post should generate some pretty healthy debate within your own mind…Cloud, Compute and Brand.
I think we’re now at a time in the evolution of compute resources where the brand of the equipment really doesn’t make any difference. Sure, there is the mean time between failures which plays a part but when components all comes from pretty much the same manufacturer, it’s really then down to assembly quality.
So what is really important? Cost and Compute Unit Measurement. How do you calculate such a thing?
In simple terms, we could look at it with this scenario. In this example we’ll use the following server specs;
- 2 RU server
- 256GB RAM
- Dual Processors with 10 cores each running at 3.0Ghz (total of 60Ghz of processing)
- Purchase price of $10,000
For this example, let’s assume you’re going to have quite a few of these servers providing compute capacity using some virtualisation technology, pick your own vendor but my preference is HyperV 🙂 But let’s also assume, you’re never going to run this server at more than 80% capacity, so the initial numbers need to be wound back to 204.8GB and 48Ghz respectively. (I honestly think you can assume you’d use 90% of the CPU capacity but your experience should tell you this)
Let’s say we’ll use a measurement of 1 CCU (Cloud Compute Unit) = 1GB of RAM and 500mhz of CPU which from my experience is a pretty good average of RAM to CPU ratio, in real life I’ve seen the CPU requirement to be a lot less.
So in this example, that piece of hardware could hold 96 CCU’s which would be totally under utilising the hardware potential but let’s go with this, as it shows a design characteristic which is required upfront.
To calculate the cost of 1 CCU per month, use the following formula;
$10,000 / 96 CCU’s / 36 months = $2.89
An important point, if you can get your RAM to CPU ratio right for the workloads you are using, you can get more CCU’s onto a single server. Drive the purchase price down and the number of CCU’s you can host on a server and your cost to provide CCU’s just keeps coming down.
There are plenty of other variables which you need to consider when calculating a CCU such as hypervisor software, network, rack space, people costs but this should be a good starting point.
So really, does the badge on the front of the piece of tin really make any difference in delivering the service?