In a recent article, Global Services asked if cloud computing [is] really green. Considering that most of your supply chain apps are moving to the cloud, and that sustainability is a must, it’s a question you should be able to answer. Unfortunately, the article was completely useless and gave you no useful information whatsoever.
All it said was that
- a recent study from Microsoft noted that, due to increased efficiency and scalability, outsourcing companies can reduce their energy use and carbon footprint by up to 90%,
- an uninterrupted power supply is required to keep data warehouses running and cool, and
- energy efficiency is not green on its own, and is no longer enough.
With respect to these statements
- we all know scale adds efficiency, and that data centres will be much more energy efficient if they use modern high-density low-power blades while you use ancient box servers,
- classic data centres will create sauna temperatures if not kept cool, and
- the energy source needs to be renewable.
In order to determine how green a data centre is, you need to know at least the following three metrics:
- The Performance Per Watt (PPW) of the hardware.
This tells you how efficient the hardware is. Typically measured in MFLOPS (Mega Floating point Operations Per Second) or LINPACK FLOPS per watt, an efficient modern IBM BladeCenter can get up to 536 MFLOPS/watt. Compare this to a desktop Beowulf cluster that maxes out at 58 MFLOPS with 4 dual core Athlon 64’s.
- The average load of each operating machine.
If a machine is only operating at 20% load, on average, you’re only getting 20% of the PPW. But if the average load of each operating machine is 80%, you’re getting at least 80% of the PPW. A modern data centre uses dynamic process allocation and virtualization and machines sleep until the load on the machines that are awake exceeds about 90% in a sustained manner, at which point another machine is woken up and added to the available pool. A traditional data centre does not use virtualization or have machines that support fast wake-up or shut-down or dynamic load balancing software and the machines are always on.
- The Power Usage Effectiveness (PUE).
PUE is a measure of how efficiently a data centre uses power and measures how much power is used by the computing equipment in contrast to the overhead required for lighting, cooling, etc. Defined as Total_Facility_Power / IT_Equipment_Power, an ideal data centre has a PUE of 1.0. While not achievable, a PUE of 1.1 is, even though the average data centre still has a PUE of 2.0 to 2.5. Today, a data centre is generally considered green if it has a PUE of < 1.5, but the best data centre in Europe, datadock, has a PUE of only 1.21 and the Thor Datacenter in Iceland has a PUE of only 1.07 and a zero carbon footprint. With carbon emission mandates coming into effect around the globe and significant mandates for power reduction by 2020, if the data centre isn’t closing on a PUE of 1.2, it’s not efficient. And unless at least one its primary energy sources is not renewable, it can not be considered green. (At the very least, the overhead should come from renewable sources.)
So don’t get confused by grandiose claims, scale, and jargon. Ask for these third-party verified metrics. Then you’ll know how green the data centre really is.