By Tony Scott

The services that you provide the business is dependent upon having a well-run data center.  The reason being is that a big part of your value proposition to the business is minimizing the overhead costs in the delivery structure.  For data centers, this means maintaining it as efficiently as possible.  These efficiencies include power, cooling, density, data throughput, and storage.  Whether the data center is owned by your company, or you co-locate with a service provider, or if you use the cloud, those costs are factored into the investment that the business has made in technology.  The costs are unavoidable, but it is how you manage those costs that will drive whether you are value add to the business or not. 

I’ve included some metrics that I’ve used over the years to measure data center efficiency.  There is much debate over which metrics should be used, especially what is presented to the business.  My advice is to look at your organization and determine which ones make more sense for you in how you deliver services to the business (for example, you will have less visibility into these metrics for a co-located solution than you would if your organization managed the data center itself).  That being said, the business does not (usually) want to see all of these metrics.  What they want to understand is that the costs are being managed ably and how those costs translate into the services being provided. 

 I’ve broken down data center efficiency metrics into these categories:

• Hardware – metrics addressing how efficient is the hardware in the data center.  This includes all server, storage, and network in the data center.  Hardware innovations and virtualization have attributed to the improvement of server efficiency.  As solid state begins to move in as a way of providing storage, those metrics will continue to improve as well. 
• Power – this is the most expensive commodity used in a data center.  Data center managers need to understand how much power is consumed, and where it was utilized.
• Personnel Requirements – measurements of how much automation has been introduced into the data center by way of how many people are needed to support and manage the data center
• Systems Management – describes the performance of the data center from an application or service point of view

One note on these metrics – when I discuss square footage, I am referring to the data center floor only (not the periphery offices that flank the data center). 

Hardware

Hardware metrics describes how much computing power can be provided given a certain amount of space.  Server, storage and network are evaluated when discussing hardware metrics.  Better utilization means lower overall costs for the data center.

• Data center density – the number of CPUs cycles (not CPUs) per square foot of data center space.  The more cycles a CPU can perform in a given time period, the better the ratio. 
• Storage density – amount of storage per square foot of data center space.  This is the sum of all storage, allocated and not allocated.        
• Storage utilization – how much of that storage is allocated to systems (whether they are using it or not).  There has been much written and debated about this metric.  My perspective has been that if the business is paying for it and it is allocated to them (regardless of whether it is actually used or not) it is unavailable for other to use, and therefore should be calculated as part of the utilization. 
• Network utilization – how much bandwidth is available in the data center and how much of it is being used.  Multiple embedded metrics included are network capacity, saturation, and availability.

Power and Heat

Power and heat have an interdependent relationship with each other.  You need power to run the data center, but that comes at a cost in the way of heat generated, which then drives cooling costs.  Also, as data center density has improved, so has increased heat generation. 

• Power efficiency – CPU cycles vs kW-h to support the data center.  That involves all costs to support the data center including storage, network, HVAC, etc.  If you have high density data center, this metric will improve.  This can go beyond just CPU cycles and encompass storage (terabytes per kW-h) and network (throughput per kW-h)
• Heat efficiency – BTUs generated against CPU cycles.  The equipment will generate heat and you have to cool that by way of power.  The more heat generated, the more it costs to supply the power.  If you have the ability to within your data center, you can extend these metrics to include storage (terabytes per BTU) and network (throughput per BTU). 
• Power Usage Effectiveness (PUE) – total power provided to the facility over the power requirements of the technology equipment
• Data Center infrastructure Efficiency (DCiE) – the power requirements of the technology equipment over the total power provided to the facility

Personnel Metrics

These metrics measure how much automation you have within the data center.  Of course the lower the count the better, but these metrics help you understand within your peer groups how efficiently you manage your data center.  Keep in mind that people in the data center contribute to degraded HVAC performance, so the fewer people you have in the data center, the better. 

• Number of employees / contractors per square foot.  It is critical that you capture all personnel resources directly responsible for the management of the data center.
• Number of employees / contracts per CPU cycle.  Another view into the management of your data center by way of CPU cycles.  I prefer to use square foot as a measure as it reflects more of the physical limitations that you are working with and it is more universally used. 
• For larger data centers, you will want to break these metrics down into these areas
  • Server and server density  
  • Storage and storage density
  • Network and network utilization
  • Facilities management by way of either of power or BTU

Systems Management

These metrics cover the services that are provided from the data center to the application teams and ultimately the business.  While I can’t say these metrics could be classified as pure “efficiency” metrics, they provide the end results of the services delivered from the data center and are important checks against the efficiencies in the data center. 

• Availability – from a data center perspective, this is a measure of the services that are available to the applications or services running inside the data center.  I have used this number as an aggregate of all of the technologies and facilities required to make a server or service available.   
• Capacity – again, from a data center perspective, how much capacity is available to the services that are being provided.  This is usually discussed as physical capacity and virtual capacity, network, storage, HVAC, and physical space. 

Other Metrics

The metrics discussed in this article deal with operational efficiency of the data center.  There are many more metrics involved in planning and managing a data center that should be considered including, emergency preparations, security controls, safety and fire controls, monitoring statistics, quality assurance, server / network / storage / software costs, performance, cooling, and layout efficiency.