Cloud Computing: Not Always the Low Cost Alternative


There is no doubt cloud computing has delivered multiple benefits to the IT organization. However, without proper management and controls, these benefits could become a non-trivial expense to the organization. In a Wall Street Journal article earlier this year The Hidden Waste and Expense of Cloud Computing, Clint Boulton outlines the pitfalls of buying too much and not tightly controlling what is bought. ISG just released a Cloud Comparison Index which is described in Stanton Jones’ blog posting and makes many of the same points.

As Boulton rightly points out, after the cloud purchase is made, another big cost management opportunity remains: managing demand and shutting down compute resources when they’re not being used. Paying for unused resources can turn a good financial decision into a bad one.

However, we’re also seeing clients jumping into on-demand cloud services assuming cloud computing is always priced better than a traditional alternative. Depending on the size of the compute segment in question, how it is going to be used and how it is purchased, a blind assumption that cloud is always cheaper, may not be valid.

Using Amazon Web Services (AWS) pricing as a comparison (AWS has had multiple reductions in their pricing and are one of the least expensive options available, hence a conservative comparison), traditionally delivered computing power can be cheaper than cloud services depending on the configuration and whether you are buying On Demand or using their Reserved pricing which requires longer term commitments.

As an example, using an 8-core, 30-gb compute resource (AWS M3.2xl) and comparing that with a traditionally delivered server with 8-core and 32-gb, the On Demand pricing becomes more expensive when you use the server more than 37% of the month. So depending on what these resources are being used for (i.e., how long they are going to be used, how much they are going to be used, etc.), it’s possible that a traditionally delivered server solution would be less expensive.

On the other hand, if you are willing to procure a traditional solution, then you are locking yourself in to that server for at least three years and if that is the case, you can make a 3-year commitment to AWS and then the pricing for AWS drops and it becomes a virtual wash (assuming the traditionally delivered server was running at maximum power utilization 50% of the time and idle 50% of the time).

Where traditionally acquired server computing still has a small advantage, is in the flexibility of when to refresh the server. The analysis above assumed a 3-year life of the traditional server to align with the 3-year term with AWS. If that same traditionally delivered server was stretched to a 4-year life, then the traditionally delivered server would be about 17% less costly compared to the AWS 3-year Reserved pricing.

Now when it comes to the small end of the computing scale (e.g., 2-core, 4-gb server), an AWS 2-core server is cheaper than a traditionally delivered alternative whether you use On Demand or Reserved pricing. The problem is most organizations won’t buy small servers anymore. Rather, they will buy a large server and virtualize it into 15 or more small instances. So, a direct comparison of a 2-core physical server to the AWS 2-core offering is not realistic. However, if your organization has a project that would require a bunch of small capacity servers for a short period of time and then they would no longer be needed, cloud would be an excellent alternative with positive economic benefits regardless of your utilization over this short period of time.

The key takeaways are that while cloud can offer pricing advantages, clients still need to align computing requirements and computing solutions and, in the case of on-demand resources, to ensure sure the actual usage is managed so the volume reflects the business case that justified the decision to go with an on-demand cloud solution.

Notes: All computations above assume: 730 hours in a month; power supplies for in house servers had max loads of 70% and idle loads of 25% of stated power supply capacity; cooling factor was 50% of electric usage; electric cost was $0.10/kWh; cost of capital for in house servers is 8%; asset life is 36 months. All in house servers were configured with no storage in the price. AWS server prices are plain Linux (not RHEL or SLES) in order to get as pure an infrastructure cost as possible.