An important element of controlling data center cost is, of course, making sure the energy use is as efficient as possible. The Power Usage Effectiveness, (PUE) is a great indicator, but it's also open to misinterpretation. Herein lies the CIO's dilemma: If improving the efficiency of your data center is an important goal, should you incentive the organization to improve PUE? In case you want to stop reading here, the answer is, "Yes, but only if you also pay attention to the bottom line."
In a recent white paper entitled "A Holistic Approach to Energy Efficiency in Data Centers" Dileep Bhandarkar of Microsoft points out that reducing fan power of your servers, for instance, would DECREASE the overall energy consumption of your data center, but INCREASE its PUE. So as a CIO, if you incentivize your data center operations on just PUE, you could end up on the wrong track.
To get the most our of PUE you need to realize two things: 1. PUE is a "system"-level indicator that provides a measure of capability only when things are optimized, and 2. Optimizing the server energy consumption of your data center still might be the right thing to do even if it raises your data center's PUE.
The big point about PUE as a "system" is reinforced in the recently published results from the datacenter2020 collaboration between Intel and T-Systems. Here incremental steps were taken in optimizing elements of the data center, all with the idea of improving overall efficiency. As the details of thewhitepaper disclose, although some of the steps taken early on (like plugging leaks in the floor tiles) had little immediate impact, once the air pressure was optimized the full benefit of that housekeeping was realized. In fact, optimizing the data center allowed a greater density of IT power consumption, further increasing the PUE.
What PUE misses is the notion of work-output of the servers and the data center. This is a generalization of the problem pointed out in the Microsoft whitepaper. PUE is a ratio of two power metrics; : like the "energy profit margin" of the data center. It's a good indicator, but just like margin, you can't run a successful business without also paying attention to the bottom line.
You can imagine having a data center full of old inefficient servers, and replacing them with the newest generation of Xeon 5600-based servers. Because of advances in power management technology, these servers will consume, even at equivalent loading, far lower power and yet will produce far greater work output.
The chart below shows the evolution of hte energy efficiency of a SP server family over the last six years. As the performance of the servers increases, their power consumption drops.
So, refreshing your servers, your power bill will go down, your facility equipment will not change, the work output of the data center will increase dramatically,and your PUE may go in teh wrong direction. Again, if the data center operations are incentivized on PUE alone, you might get the wrong behavior.
So as CIO you face a dilemma. You want to increase your data center's energy efficiency. You have heard about PUE? What do you do? The right thing, of course, is , "Pay attention to PUE, but also pay attention to the bottom line indicators like Power Consumption and Power Costs."
Of course, the real issue is optimizing the productivity of the Data Center. For highly homgeneous workloads, I imagine the metrics are well established. But I think there is still tremendous work to do in defining more general productivity indicators for data centers.
What do you think?