Published:
January 05 2010, 11:05 AM
|
1 Comment(s)
by
Dhesikan Ananchaperumal
Did you know that implementing an energy efficiency project can have an adverse effect on your energy efficiency metrics? Many organizations today are taking a hard look at their energy efficiency and developing programs to improve on their stats. Many of these initial projects are taking place within the datacenter – a large consumer of energy (and high energy cost center) within most enterprises. If you’re relying on PUE or similar metrics to track your progress, you might be surprised at the results.
As you know, PUE is inversely proportional to the power consumed by the IT infrastructure. And our goal is to keep the PUE as low as possible for greater efficiency. A datacenter with a PUE of 1.2 is using 83.33% of the incoming power for running IT infrastructure related load. According to a recent Virtualization Journal post, “Most data centers today consider a target value of 1.5 good, with some companies such as Google trying to drive their PUE below 1.2 – an industry benchmark.” There’s always room for improvement, so now, suppose the IT team is planning to implement a virtualization project that will reduce the number of physical servers in a 5:1 ratio.
The goal is to reduce power consumption and total space needed for the devices, and to improve heat dissipation. The IT team decreased the number of server/network-gear racks from 5 to 1, which is approximately an 80% reduction in physical space, and as a result, reduced the server/network-gear power consumption. By improving circulation, they also reduced the amount of cooling required.
All told, let’s say this project resulted in a 50% reduction of power consumed by IT load – a respectable result. Now, what happens to our overall PUE value? Before the virtualization project, the datacenter’s PUE was 1.2. If you do the math, after implementing the virtualization project, the PUE has gone up to 1.4!
One way to interpret these numbers is to assume that the overall energy efficiency went down. But that is not true. Another interpretation is that we need some energy optimization on the Non-IT load side, bringing the PUE value back to 1.2 or less. Based on the business needs and infrastructure set up, we might be able to come up with some user specific formulas to provide more clarity to these variations.
Herein lies the rub when it comes to interpreting energy efficiency metrics. Although we did a project to increase energy efficiency within our datacenter, the overall PUE metric shows the opposite results. How do we make sense of this?
Overcoming this discrepancy is an area that is still being worked on. It involves improving the definition of energy efficiency metrics like the PUE. One approach, in my view, is to associate the PUE variation to the energy and environmental variables within a datacenter. For example, we can have a PUE variance chart/graph with different function definitions for PUE variance. If we consider PUE variance as a function of outside air temperature, then the PUE variance chart/graph will provide the various details on what the actual PUE value should be depending on the outside temperature at any instance. A template of such visualization may look like the one below.
If we do a trend analysis using any type of regression, then the following may be one of them in which the regression type is linear.
In our previous scenario, where the PUE went from 1.2 to 1.4 after implementing an energy saver virtualization project, we should consider PUE variance as a function of the IT load. By doing this, the PUE variance chart/graph will clearly provide the necessary justification for the increase in the PUE value.
Other ideas that are being tossed around for this is to use the old performance per watt kind of metrics to understand datacenter energy efficiency. If we can minimize the IT load by using virtualization or cloud computing, then the associated facilities overhead will also be minimized. Also, there are some concepts on DCP Index (Data Center Productivity Index) in which we define the useful work within a datacenter which is a challenge as well.
Although energy efficiency within the datacenter is a well known area of interest to many of us, it’s clear that there are some unresolved challenges involved with understanding the meaning of available metrics and learning how best to interpret the resulting data. Depending on the business needs, infrastructure set-up, geographical locations, etc. of a datacenter, both the efficiency metrics being used and the relationship of these metrics to different variables may differ, and even slight changes to any of the variables can lead to misleading results. While clearer guidelines and best practices need to be developed within our industry, in the meantime, it’s important that each organization come up with its own functional definition to be able to show best results for improving the overall energy efficiency.
By: Dhesikan Ananchaperumal
Dhesikan Ananchaperumal is a Vice President of CA’s ecoSoftware business unit focusing on energy and sustainability management. He is responsible for the overall strategy and approach, product management, development, quality assurance, customer implementations, and supporting engineering. He began his...
Read More..