Allocating data centre energy costs and carbon to IT users

When a company sets out to reduce data centre energy costs and minimise adverse societal or economic costs, it needs to make energy management a priority. The Catch 22 is that expense and complexity escalates and ROI declines when excessive precision is specified. So the question is raised as to how the energy management process can be simplified and the number of physical measurements reduced so that energy use and carbon can be allocated to IT users.

The following outline shows how to overcome the barriers to an effective energy management programme easily, quickly and with a reasonable degree of accuracy. Most importantly, this can be implemented at low cost and with minimum instrumentation.

Goal Setting

There are generally three goals for a system for assessing data centre energy efficiency or carbon impacts:

• One-time or periodic performance benchmarking:
  To establish the need for an energy management programme, will not provide actionable information to drive reductions, but rather reveals performance signature in comparison to similar facilities.
• Pass-through allocation of energy or carbon:
  The objective is to incentivise behavioural changes to reduce unnecessary energy use and carbon emissions.
• Use of information to reduce infrastructure energy use or carbon impacts:
  The objective is to provide the data necessary to identify and quantify opportunities to change equipment, configurations and settings that would reduce energy consumption without an adverse effect on the IT load.

Clearly a combination of goals is the best way forward, but should that entail a process which is large, complex and expensive?

Measure or Model?

Most energy management discussion centres on metering or measurement of energy. But any rational approach requires a level of interpretation so that the effects of changes and their impact on energy use can be understood. Therefore, before we start, we need a model of how the data centre works for while the power consumption of servers can be measured directly and conceivably associated with IT users, the majority of data centre power is used by loads other than the servers.

A good model will incorporate all of the goals described above. Measurements without a model are of little value. Models, however simple, may have considerable value even with incomplete measurements. It is the model which supplies actionable information to effect change in data centre efficiency. If a perfect model could be created, there would be no need for any measurements; it would incorporate the characteristics and operating conditions of all infrastructure equipment, it would accept as inputs historical weather data and simply compute energy flows. However, we don't inhabit a perfect world.

However, a surprisingly good model can be created using only a rough inventory of the infrastructure and IT equipment deployed, some information about configuration and system design and some basic knowledge about the electrical characteristics about the IT and infrastructure.

What should be measured?

The accuracy and frequency of measurement in a management system have major implications on complexity and cost. At one extreme, it is possible to instrument every device in the system and collect accurate data about energy use, 24x7. However such a method would no doubt be uneconomic and impractical from many points of view, not least of which the power it would demand in data analysis and storage. At the other extreme, we could consider making no measurements at all.

With a no measurement system, energy and carbon costs could simply be allocated to users on a "per-average-server" basis, but the accuracy would only be ±36%. While this may provide useful information to guide behavioural change, it would not be helpful for making improvements to power or cooling infrastructure. Overall, however, it provides remarkable benefit to any manager and since it can be obtained for little or no cost, it should given high consideration by anyone who wants to start controlling energy costs immediately, but little time or resource.

Clearly the best approach will be somewhere in the middle - a "good enough" energy data collection system combining enough accuracy to secure management goals with low cost and high ROI. Such an approach brings together modelling with some measurement. It suggests that a no-cost system of energy modelling based on server counts, UPS power readings and approximate inventory is good enough to allow meaningful energy allocation to IT users. Over time, more detail can be added to improve the model and mitigate error. A full explanation, together with illustrations of the various approaches can be found in APC White Paper #161; "Allocating Data Center Energy Costs and Carbon to IT Users".

Allocating Energy to IT Users

Although IT capacity can be measured in a variety of ways, one simple method is to count the number of servers. If each IT user can be allocated a number of servers, then all that is needed is to attach an energy and carbon value. This requires the identification of all energy uses in the data centre and their allocation on a per-server basis. It includes the server's own energy use, coupled with that for lighting, storage, networking, power, cooling and auxiliary loads.

This is highly simplistic and prone to be inaccurate, however, users can develop a power profile for each server type. Each server then has a base standard power level assigned plus an allocation representing a fraction of the overhead power for all supporting networking, storage and power devices and infrastructure.

Having allocated energy use to IT users, it is relatively straightforward to calculate the carbon emissions created by each user.

Conclusion

No-cost models can be applied effectively and logically for energy management. Although such a model may not be precise, it is sufficiently accurate to inform decision making and it can be enhanced over time as more accurate inputs are included to improve the model. It should not be assumed that complex and extensive metering is a base requirement, in fact it can provide a poor return on investment. It should be noted that every wasted watt is an unrecoverable loss and that even a simple energy management system such as that described can be an important step to reducing data centre energy use and carbon footprint. For more details about the methodology, please download White Paper 161: "Allocating Data Centre Energy Costs and Carbon To IT Users", available from www.apc.com.