According to the EPA’s 2007 report to Congress on server and data center energy efficiency,the two largest consumers of electricity in the data center are the support systems (50% of total) and general servers (34%). Last week I wrote about Energy Star Servers which over time will drive down the electrical demands of the servers. Today I’d like to focus on overall data center efficiency and productivity and the efforts of the Green Grid industry alliance.
For the most part, being green in the data center is about increasing power efficiency and thereby reducing greenhouse gas emissions. From a purely environmental and ecological perspective, higher levels of green could be achieved by examining the entire life cycle of technology systems— from production and supply chain all the way to the disposal of toxic components. But for now let’s stick with the power efficiency theme.
Here are some Green Grid developed metrics for evaluating data center facility efficiency:
- PUE—Power Usage Effectiveness represents how much total power you need to drive your IT systems when you take into account power distribution, cooling, humidification, lighting, etc. If you need 1MW to run your IT systems, a PUE of 1.8 means that you are consuming 1.8MW to get this usable level of power.
- DCiE—Data Center Infrastructure Efficiency represents the ratio of IT equipment power to total facility power. Using the same example, a facility with a PUE of 1.8 has a DCiE of about 55%, meaning a little more than half of the power used by the data center is making its way to the IT systems (no mystery here; DCiE is simply the inverse of PUE).
From a planning perspective, I’ve personally found PUE to be a more useful tool than DCiE because it gives you a simple multiplier you can use to estimate your increasing facility requirements as you add new servers, storage and the like. While PUEs generally fall into the 1.7 (more efficient) to 3.0 range (less efficient), I have seen PUEs approach 1.3 when a holistic, facility-wide approach is taken regarding power efficiency.
Keep in mind, the PUE and DCiE numbers tell you how efficiently your data center is operating from a power distribution and cooling perspective. They do not tell you how efficiently your IT group is delivering useful value per kW of total power consumed.
This is exactly what Green Grid is trying to do next: define ways to measure overall Data Center Productivity (DCP). That is, Green Grid is attempting to help you quantify how well the energy you’re using is being applied to useful work. Since this is very complicated and will vary by application and industry, the Grid is hoping to define useful “proxies” or indicators that can provide some normalized measurement and useful insight.
Here are some of the types of measurement proxies being considered by the Green Grid (read the Green Grid Proxy White Paper for more details):
- Server productivity.Because you need a greater number of older servers to get the job done, and because older servers consume a lot more electricity per instruction executed than newer ones, you could come up with a productivity measurement based on MCUPS (million compute units per second) per kW consumed by the data center (the more, the better). This, of course, ignores the value of stored data which may be considerable, or even required for regulatory compliance. However, it provides a measure worth consideration.
- Server utilization and virtualization. Running high-end servers at 20% capacity means they are being used inefficiently.The idea here is that running servers at higher utilization rates is good and a highly virtualized environment has the
- (server utilization continued) potential to produce more useful work per kWh consumed. Of course, this would not be good way to measure efficiency for applications that are seasonal or “bursty”— like those used by retailers (for, say, the Christmas rush), financial brokerages (with sudden floods of transactions) and even national security applications that are built for peak loads to guarantee performance in any eventuality.
- Bits out per kWh. The idea here is that it may be possible to measure useful work simply based on output. This may in fact be a great proxy for data intensive applications like websites, VoIP and video where the name of the game is moving bits efficiently. On the other hand, this would be a poor measurement for applications that aggregate and analyze data, then report out digestible, concise information, such as in business intelligence applications.
My purpose here is not to disparage the efforts of the Green Grid regarding DCP. I do believe, however, that the DCP models could use more input from the user community. That’s right: I mean you. Please post any comments here or, if you prefer, you can get involved with the Green Grid directly and provide feedback to the DCP survey.
At the same time, I encourage you to not lose sight that optimizing your PUE offers low hanging fruit in achieving improved efficiency, reducing carbon emissions and saving money. Modern facility technologies such as ultrasonic humidification, high efficiency harmonic mitigating transformers and variable frequency drives (VFDs) can be effective in dropping your PUE into the more efficient 1.3-1.5 range.
In fact, let’s compare IT server and storage upgrades with facility infrastructure improvements (PUE):
- For a baseline, let’s assume your IT equipment consumes 2MW with a PUE of 2.5. By investing in more efficient servers and storage, let’s say you are able to cut IT power demand by 30%. If the cooling and power systems scale down as well, you will cut your total power from 5MW to 3.5MW. Not bad— but also not really a quick fix as you migrate apps to your new servers and retire old systems.
- Now, using the same scenario, let’s say we improve facility efficiency instead (the support systems—cooling, humidification, distribution, etc.), driving PUE from 2.5 to 1.5 with changes that are invisible to the applications, servers and storage systems. In this scenario, you could reduce total electricity by 40% to 3MW and do it with a lot less migration expense and pain.
There’s no doubt that replacing older,inefficient computers and storage systems with more energy efficient models is a good idea. Clearly, ENERGY STAR servers and storage will help through natural technology refresh cycles. Virtualization will reduce the number of servers you need, allowing your systems to run at higher operational efficiency.
However, if you want to make an immediate impact that will continue to pay dividends as you refresh your servers, storage and network systems over time, improve your PUE first.
As always, thank you for sending comments, tips and topic suggestions to me at CIOblog@TransitionalData.com.
Michael Bullock is the founder and CEO of Transitional Data Services (TDS), a consulting firm helping clients implement energy saving green data center solutions, data center relocations, web based enterprise applications and 24/7 technical operations.