17-6-8: Data Center Efficiency: If you can’t measure it, you can’t improve it
Over the past few years, efficiency has become the center of attention in data center facilities resulting from the increasing need for computation. It is important to note that this increase in energy consumption by servers has impacted not just the cost of energy used but also the cost of the infrastructure to support the server. In fact, in many cases, these costs have already exceeded the cost of the server.
As a result of these and other factors, energy efficiency has become extremely important to the industry. This is evidenced by the fact the many forums and activities around efficiency have emerged. Some of the more notable:
The formation of The Green Grid - This consortium of companies was formed to develop and promote the adoption of energy efficient standards, processes, measurements and technologies
Industry white paper for developing an efficiency metrics for servers;
President Bush signed the bill that will authorize the Environmental Protection Agency (EPA) to study efficiency in IT equipment and data centers. The EPA is now in the process of drafting a report to recommend what actions the government should take to curb the rising power consumption of data centers.
However, while much attention has been given to servers, the truth is that their efficiency has been drastically improving year over year. Look at the relative performance and performance/watt improvements of a typical server since 1999. In this example, indicative of general server performance trends, performance has increased by approximately 75 times between 1999 and late 2006. During that same period, the performance per watt has increased 16 times, or essentially doubling every two years. This indicates that servers are becoming significantly more efficient year over year. For the automotive industry to keep pace with this efficiency improvement, a car getting 20 mpg in 1999 would get about 320 mpg today.
If this efficiency improvement is true, why has the demand for IT equipment exploded? The reason: The insatiable demand for compute cycles is spurred by the ever-decreasing cost of computation and the build-out of the internet infrastructure. As the cost goes down per compute cycle, the demand goes up and more applications emerge that have an acceptable ROI. The issue though is that with this growth in computation comes data center power and cooling challenges as customers add denser, higher-power equipment to meet the compute demand. Recently, Gartner predicted that half of the world's data centers will run out of power by the end of 2008. As a result, industry is aggressively advocating efficiency metrics such as the industry white paper to set the foundation for a server metric to promote behaviors that improve efficiency even further. Similarly, there needs to be industry-wide efficiency metric for data centers.
Data center efficiency metric
As new technologies emerge in the data center that claim improved efficiency, there needs to be a way to measure whether there is improvement. Currently, all efficiency claims with data center infrastructure technologies are based on hearsay and marketing "hype". There needs to be a fundamental way to benchmark data centers and compare them over time. A good metric would help answer the following questions:
How well are you running your data center relative to others in the industry?
Are your new data centers more efficient then the old ones?
Which data center should you obsolete?
Does implementing best practices really improve efficiency?
When you adopt new technologies, do they actually do what your suppliers promised?
Ultimately, just as every car and efficiency rating e.g., MPG in the US, we should have a rating for data centers. An efficiency metric called Power Usage Effectiveness (PUE) was first introduced in 2006. Since then, the Green Grid has adopted it as a valid metric to measure data center efficiency and has published a white paper recommending that PUE be measured in all data centers. This metric is defined as the ratio of the total facility power in the data center over the power of the IT equipment on the raised floor:
The total facility power is the power delivered to operate the data center, including power for operating the IT equipment and all of the power and cooling infrastructure. This includes the power for the IT equipment, switch gear, Uninterruptible Power Supply (UPS), chiller, cooling tower, air conditioners, liquid conditioners, etc. The IT equipment power is defined as the actual line cord power drawn by all the IT equipment in the data center. It does not include the power losses associated with conditioning and reducing the voltage from the utility. The intent is to consider only the power used for IT equipment.
PUE is useful for understanding the total amount of energy consumed by IT equipment relative to the total power. The metric captures the actual cost to the business at the utility meter for a watt of IT equipment power. The annual energy cost for a server, storage unit, etc., is determined using the following equation:
Annual Energy Use Cost = (8760 hrs/yr) x (Average Utility Rate in €/kW-hr) x (Equipment Power in kW) x PUE (2)
The importance of this metric is highlighted by the Green Grid's adoption of PUE as the efficiency metric that the industry should start using for both benchmarking purposes as well as calculating energy costs as shown in equation.
Introducing Compute Power Efficiency (CPE)
While PUE captures data center efficiency, there still needs to be a way to capture how efficiently the data center power is used for computation. To capture the computational efficiency, a new metric is introduced called Compute Power Efficiency (CPE). CPE is defined as:
CPE = IT Equipment Utilization / PUE
= IT Equipment Utilization * IT Equipment Power / Total Facility Power
This metric provides a view of what percentage of the total facility power is actually used for computation. It connects IT hardware efficiency with the data center infrastructure efficiency. The metric may be applied to an individual server or an entire data center, provided a weighted average utilization may be determined. In many data centers, enterprise server utilization is typically 20% or less. This means that for a typical, well-managed data center with a PUE of 2.0, the CPE is 10% or 1W for every 10W of utility power is actually used for computing. Data centers with a PUE of 3.0 have a CPE of 6.7%. Our preliminary analysis using CPE indicates that a tremendous amount of energy is wasted in over-provisioning compute and cooling resources in most data centers, and will be outlined in more detail in the future. To fully address the cost of an IT operation CPE must be optimized.
Conclusion
PUE is a metric that can help benchmark improvements in data center operations. Thus, it is hoped that there will be broad industry adoption. With broad adoption and the support of government agencies such as the EPA and the European commission, the hope is that we can already capture PUEs around world and develop a comprehensive data base. Once this is done, we can really understand the good, the bad and the ugly in data center design, implementation and operation. Then we can measure the efficiency and improve it.
Finally, a promising metric that needs further development and discussion was introduced called Compute Power Efficiency (CPE). With more work, this metric could be useful for IT mangers seeking to optimize both power and computational efficiency in the data center.
|
