Slash data-center costs and downtime by using Coolan's TCO Model

What is the most cost effective way to meet your data-center needs: in-house, colocation, or managed services? Coolan's open source solution can help you answer this question.

Image: iStock/scanrail

At the 2015 summit of the Open Compute Project (OCP), Amir Michael, a well-respected data-center engineer and a founder of OCP, offered a surprise during his presentation. "Something not often talked about in the industry," Michael mentions. "There is a trend of people moving off the cloud."

Amir Michael
Image courtesy of Coolan

How Michael came to this conclusion is an interesting story. Michael, his brother Yoni Michael, and Jonathan Heiliger, all cofounders of the startup Coolan, developed an analytics platform called Coolan. The purpose of Coolan is to help data-center operators reduce downtime and lower infrastructure costs by analyzing metadata on how the servers are performing. During Coolan pilot programs, Michael noticed participants were in a quandary about where to place their data-processing and data-storage needs.

"Trying to figure out the TCO (Total-Cost-of-Ownership) for your compute infrastructure is no easy task," explains Michael. "With the proliferation of public, private, and hybrid clouds, not to mention the ever-growing number of Buzzword-as-a-Service paradigms, anyone in charge of their company's infrastructure faces a difficult decision: Should you build or acquire your own data center, lease space in a co-located facility, or rent a piece of the cloud?"

Coolan's answer to the infrastructure TCO question

When I interviewed Michael about Coolan, he mentioned that Coolan was an interesting and challenging project. It seems Michael likes that kind of challenge, a little before and during Coolan's development, Michael along with Jimmy Clidaras, a distinguished engineer for data center and platforms infrastructure at Google, were working on a side project that specifically addressed a data-center's TCO.

Jimmy Clidaras
Image courtesy of Coolan

Three months later, the two had what they call the TCO Model. It may look like a simple spreadsheet (a populated example and a working blank spreadsheet), but I am assured there is a great deal of analysis going on in the background -- the results of hours and hours of research by the team. The screen shot in Figure A is just one of the 11 pages in the spreadsheet.

Michael mentions, "We built the TCO Model to let operators evaluate and understand the cost points for different types of deployments."

Some of the cost points the TCO Model addresses include:

  • Whether to build your own servers or buy OEM
  • What type of hardware components to use
  • Whether to upgrade systems or deploy new ones
  • The cost of capital
  • The data-center PUE
  • The amount of power the network is consuming

And the ultimate question: stay in the cloud, go colo, or build a private data center?

Figure A

TCO Model
Image courtesy of Coolan

Findings of interest

During his presentation at the OCP summit, Michael talked about some of the more interesting results the team discovered.

The cost of power was one area the team looked at closely -- is it a significant expense? To their surprise the answer was no. Michael said the cost of electricity for a customized in-house data center was only 3% of the overall cost. Electricity amounted to 5% of a colocation data center's total costs (Figure B).

Figure B

Image courtesy of Coolan

Michael also notes in the talk that what is considered standard methodology isn't always appropriate. Each data-center operation is unique. In one sample case, Michael arrived at the following:

  • Managed services in the cloud wins if a company requires more computing power than storage.
  • In a related sense, the amount of storage required has a huge impact on choosing a deployment method.
  • The cost to grow is less for in-house data centers than cloud-managed services or leasing more colocation space.

The conclusions may seem to go against current thinking, but accurately modeled the TCO for this particular data center.

Clients are customizing the TCO Model

Michael is excited that clients are finding interesting ways to plug data into the model, permitting them to figure out things they did not know before. In the end, Michael and crew hope that by offering the TCO Model as an open-source (free) platform, more transparency will be added to the industry, allowing data-center operators to understand their costs, which will in turn allow them to make better decisions.

Also read