The Shift to a Cloud Cost Model - A money saving approach

As you know, you pay for resources in the cloud as you use them. For Amazon, that model is by the CPU-hour. For other clouds, such as GoGrid, it’s by the RAM hour. Let’s look at how you can anticipate costs using the example resource demands by using (two application servers from midnight until 9 a.m., eight from 9 a.m. until 5 p.m., and four from 5 p.m. until midnight)

Suppose your core infrastructure is:

• $0.10/CPU-hour: one load balancer
• $0.40/CPU-hour: two application servers
• $0.80/CPU-hour: two database servers

Each day you would pay:

$2.40 + $44.00 + $38.40 = $84.80

Your annual hosting costs would come to $30,952.00—not including software licensing fees, cloud infrastructure management tools, or labor.

How to Approach Cost Comparisons

The best way to compare costs in the cloud to other models is to determine the total cost of ownership over the course of your hardware depreciation period. Depending on the organization, a hardware depreciation period is generally two or three years. To get an accurate picture of your total cost of ownership for a cloud environment, you must consider the following cost elements:

•  Estimated costs of virtual server usage over three years.
•  Estimated licensing fees to support virtual server usage over three years.
•  Estimated costs for cloud infrastructure management tools, if any, over three years.
•   Estimated labor costs for creating machine images, managing the infrastructure, and responding to issues over three years.
•  Any third-party setup costs for the environment.
  

If you want a truly accurate assessment of the three-year cost, you will need to take into account when the costs are incurred during the three-year period and adjust using your organization’s cost of capital. This financial mumbo jumbo is necessary only if you are comparing cloud costs against a substantial up-front investment in a purchased infrastructure, but if you understand it, it’s still a good idea.

In comparison, you must examine the following elements of your alternatives:

• What are your up-front costs (setup fees, physical space investment, hardware purchases, license fee purchases)?
• What labor is required to set up the infrastructure?
• What are the costs associated with running the infrastructure (hosting space, electricity, insurance)?
• What are the labor costs associated with supporting the hardware and network infrastructure?
• What are the ongoing license subscription/upgrade fees? Maintenance fees?

 
This particular example assumes that two application servers easily support standard demand. It also assumes, however, that the business has a peak period of traffic on the 15th day of each month that lasts for 24 hours. Serving this peak capacity at the same performance levels as standard capacity requires an extra four servers. This system can still function at peak capacity with only two extra servers—but at the expense of degraded performance.

If you’re starting from scratch, you will minimally need the following equipment for your IT shop:

1. Half rack at a reliable ISP with sufficient bandwidth to support your needs
2. Two good firewalls
3. One hardware load balancer
4. Two good GB Ethernet switches
5. Six solid, commodity business servers (we will accept degraded performance on the 15th day)

For the cloud option, you will need just a few virtual instances:

1. One medium 32-bit instance
2. Four large 64-bit instances during standard usage, scaled to meet peak demand at 8
   

In addition, you need software and services. Assuming an entirely open source environment, your software and services costs will consist of time to set up the environments, monitoring services, support contracts, and labor for managing the environment. Table below lays out all of the expected up-front and ongoing costs.

For each option, calculate the present value of your monthly payments and add in any up-front costs:

1. Internal: = (–PV(10%/12,36,1900,0)) + 53200 = $112,083.34
2. Cloud: = (–PV(10%/12,36,3900,0)) + 1200 = $94,452.63

Not only is the cloud cheaper, but the payment structure of the cloud versus up-front investment saves you several thousand dollars.

The final point to note is that you can use the $112,083.34 and $94,452.63 numbers to help you understand how much money these systems need to generate over three years in order to be profitable (make sure that calculation is also using today’s dollars).

Where the Cloud Saves Money

Cost savings in the cloud become significant, and even reach absurd levels, as your variance increases between peak and average capacity and between average and low capacity.

My company has an extreme example of a customer that has very few unique visitors each day for most of the year. For about 15 minutes each quarter, however, they have the equivalent of nearly 10 million unique visitors each month hitting the website.

Obviously, purchasing hardware to support their peak usage (in other words, for 1 hour per year) would be insanely wasteful. Nevertheless, they don’t have the option of operating at a degraded level for that one hour each year.

The cloud is the perfect solution for them, as they can operate with a very baseline infrastructure for most of the year and scale up for a few minutes each quarter to support their peak demand.

A common and relatively straightforward set of cost savings lies in the management of nonproduction environments—staging, testing, development, etc. An organization generally requires these environments to be up at certain points in the application development cycle and then torn down again.

Furthermore, testing requirements may demand a full duplication of the production environment. In the cloud, you can replicate your entire production infrastructure for a few days of testing and then turn it off.