PUE and Demand Reduction Using Solar and Ice at AFCOM Data Center World

At the AFCOM Data Center World conference in Nashville, TN.  I’m actually presenting tomorrow on utility incentives for data centers, and am looking forward to that.

There’s been a good facilities greening track that covers a lot of issues related to data center efficiency.  The most popular topic I’ve seen is about Power usage effectiveness (PUE). If you haven’t heard about PUE it’s time to study up because I think it’s here to stay.  Popularized by the Green Grid, PUE is a measure of the overhead of the infrastructure of a data center (technically it’s the power of the whole facility divided by the utility IT power).  So a high PUE means you spend more power than you need providing power and cooling to your IT equipment, or that your infrastructure is less efficient.  The average PUE according to the EPA’s latest data collection is about 1.9, meaning facilities are spending almost as much energy powering and cooling the IT equipment than the IT equipment uses itself.

As an aside PUE’s prime competitor was DCIE, which was the inverse, and so measured as a percentage that reads more like an efficiency metric.  However, PUE won the day because it was believed a metric expressing overhead would be more digestible by the C-suite.

Here’s a great slide describing what PUE is from a presentation from Steven Carlini of APC on PUE “hype”.

It seems that PUE is everywhere and people are giving advice about where and how to measure it, what it means, and how different decisions in the data center affect the PUE.  It’s great that PUE is taking hold because it will lead to greater instrumentation in the data center and is a starting point to talk about the facility infrastructure efficiency. The EPA’s uptake of PUE for their data center building rating (taking effect in June) will also help standardize the way people measure and report this metric. That said, PUE is not without its problems. The APC presentation did a great job of explaining a lot of the drawbacks of PUE, including how not all measurements of PUE are created equal so you need to make sure you’re comparing apples to apples (the Green Grid is working on making the metric more comparable across facilities).  In addition, certain improvements to your IT load can actually increase your PUE and make the infrastructure look less efficient if you don’t appropriately scale your power and cooling subsystems.

All in all, PUE is great for the industry and it will get even more useful when it becomes more standardized in how its measured and reported.  For now – it’s a useful tool for incremental improvements to a facility, but make sure you know what’s behind the numbers.  PUE is a useful metric, but the most important is overall energy saved (or demand reduced) by the facility.

I also saw a great presentation from i/o Data Centers, which I mentioned before in a post about demand reduction due to their peak shaving system of creating ice at night to cool the data center during the day.  The speaker kept saying “it’s all about power” and that space isn’t the prime issue anymore.  They’re also now planning an 11-acre solar array on top of the same data center, partially to lower their utility feed usage during peak times.  Here’s an article on that from Data Center Knowledge on how the ice system and solar work hand in hand.  Pretty cool and innovative stuff.

Demand Response (DR) for Reduced Peak Power in Data Centers

One interesting approach to demand reduction is the idea or demand response, or “DR” programs.  The New York Times recently had this article on Idaho Power’s approach to DR.  The article includes this explanation of what DR is:

This concept, called demand response, has gained traction in utility circles. In essence, it involves paying users to make small sacrifices when there is an urgent need for extra power (the “peak”). The utility can then rely on cutting some demand on its system at crucial times — and, in theory, avoid the cost of building a new plant just to meet those peak needs.

There are many opportunities for demand response in data centers. EMI did a process evaluation for the California Emerging Technologies Program (ETP).  During this project, EMI prepared a number of case studies on different technologies assessed by the ETP. One such case study was on an “Auto-DR” technology.  My colleague who worked on this passed on this report on a joint effort between PG&E and LBNL’s Demand Response Research Center (DRRC) on an a similar Auto-DR pilot program in the summer of 2006. During the pilot program, they setup locally participating businesses to have automated controls to lower their energy consumption in response to demand response signals from PG&E. Of the 24 facilities that participated in the pilot, an office/data center had highest achieved demand reduction for a single event at 363 kW and highest average for 294 kW. In this instance the the DR strategies used at the data center site included: duct static pressure increase, Supply Air Temp (SAT) increase, fan VFD limit, chilled water (CHW) temp increase, and cooling valve limit. The chart below from the report shows how high the demand savings was for the office/data center (all the way on the left) compared to other sites.

The office data center also had the lowest payback period at 0.4 years for implementing the Auto-DR.

Following the project, the DRRC published this data sheet with information on the DR potential of data centers.  The sheet makes some interesting points including that “savings can be higher than those in other industries because reducing server loads simultaneously reduces cooling and other equipment loads.”

Here are some of the other methods the DRRC recommends in their fact sheet:

–      Dynamically shift load onto fewer servers using virtualization.

–      Migration of load to another location (i.e. another data center).

–      Temporarily raise set-point temperatures.

–      Use backup reserves such as ice storage or chilled-water storage for cooling.

PG&E is still running the Auto-DR program along with the other large California IOUs which also have programs.

“Energy Savings” versus “Demand Reduction”

A funny thing in the efficiency and utility segment is the constant confusion between power and energy, or between kilowatts and kilowatt-hours. Even among engineers who clearly know the difference, it’s interesting that people are still constantly confusing the two, or at least use them interchangeably in situations where you really can’t.  A sure way to insult an energy engineer is to say they “can’t tell a kW from a kWh.

Technically, power (kW) is an instantaneous measure of the rate in which you’re using energy, while energy is a cumulative measure of how much of a resource you’re using.  Confused?!

Another way to think of it is that our resources, such as coal or natural gas, store a certain amount of energy. The power you draw dictates the rate at which you are using this energy.  If you draw more power you’ll use up your coal more quickly.  If you use less power your coal will last longer.  Obviously using less power is a good thing because your resources will last longer.

Turns out this distinction between power and energy is very important in the utility industry, and therefore effects how they run their conservation programs.  As residential customers, most people are used to being charged by the kWh, a measure of cumulative energy used.  A residential utility meter therefore measures the cumulative kWh you use in a given month, which is what the utility charges you for.

One of the chief goals of conservation programs is to reduce the peak power draw (kW) on the system, and not necessarily the overall energy used.  The peak demand is what dictates how many power plants need to be running to service a population.  Utilities that are trying to delay the construction of new power plants will look to reduce the peak power demand from their customers.  For this reason, large energy users (such as commercial and industrial customers) will pay not only for the energy usage, but also for their peak demand usage.  This gives a clear pricing signal to the customer to reduce peak demand.

A lot of utility energy efficiency programs will focus on reducing demand and will pay incentives based on reductions in peak kW – not kWh savings.  Austin Energy and Southern California Edison are two examples I’ve found of utilities that base some incentives on kW reduction.  Often this reduction needs to happen during times of peak demand to be eligible for incentives. Typically peak demand occurs in the middle of the afternoon on a hot summer day when everyone is running the AC. In contrast, here in the Pacific Northwest almost no one has AC but most people have electric heat, so demand peaks on cold winter days.

To illustrate this concept, here’s a peak load curve on a natural gas plant I found at natgas.info.

This topic came to mind as I’ve been preparing my materials for my talk at AFCOM, where I will be summarizing data center utility incentives to data center managers and IT professionals. I’m seeking different ways to explain these concepts in simplified terms so that industry members can understand the utilities’ motivation in running conservation programs.

This whole concept reminded me of a neat article I read a while back about an Arizona data center taking advantage of time-of-use pricing.  The data center has installed a system that makes ice at night and uses that ice during the day to cool the data center.  While systems like this don’t necessarily produce energy savings, they create a significant amount of peak demand reduction by shifting that demand to off peak hours (this approach is often referred to as “load shifting”).  The local utility (Arizona Public Service Co.) charges only 2¢ per kWh for off peak energy and 13¢ per kWh for on peak energy.  In this way, the company can save 11¢ per kWh (or 85%) on their energy costs by shifting the demand to off-peak hours (this is probably not quite true since there are likely some loses associated with the load shifting but you get the point).

One of the attractive things about the “smart grid” or “smart meters” is the ability for utilities to offer different prices based on time of use to residential customers so that consumers can reduce their individual peak demand.  This would help further flatten the load profile, reduce the number of power plants needed to service the population and help make electricity production more efficient.

I’m sure that I, like others, will continue to interchange the words energy and power when talking about conservation, but it’s often important to recognize that difference when identifying opportunities to increase efficiency in the system.