How Much Does Economizing help? by Peter Judge

Views and Opinions on Green IT (Feb 28 2011)

1. 

   How far is the current wave of “economising” in data centers going to get us in reducing energy use - or at least making us use our energy more efficiently?

   “We’re starting to reach the limits of what we can do,” Liam Newcombe of Romonet told me last week. 

   Newcombe is one of the people involved in the EU Code of Conduct for data centres, and the British Computer Society’s data centre specialist group. When he’s reaching limits, the rest of us probably still have a little way to go.

   But if he is reaching limits, what does that mean for data centres.

   In one sense, he has a very interesting point - we may actually be doing better than we might think we are in tuning up our data centers.

   Power waste - getting beyond the chillers

   He showed me graphs of where the power goes in a 500kW data centre, with a not especially impressive PUE (Power Usage Effectiveness) of 1,7. For every Watt sent to the IT equipment, that building spends another 0.7W, which doesn’t sound very good (and can easily be bettered in pretty much any data centre). But even so, the IT equipment gets 3.8MWh a year and no other single part of the data center gets much more than a quarter of that.

   The “wasted” energy is distributed between a whole bunch of other uses - and that’s one reason it may be difficult to make massive strides in power efficiency.

   The chillers and the computer room air conditioning (CRAC) system each use about 1MWh a year of energy, so they are the focus for efficiency moves.

   Simply containing the cold air better, and stopping warm air recirculating means that the inlet air doesn’t have to be kept ultra-cold, and putting variable speed fans on the CRAC units means they can use less power when less cooling is needed.

   These two measures could take the PUE to 1.3, without any major changes. But at that point, none of the places where energy is wasted is as large as the (pretty-much irreducible) uninterruptible power supply (UPS) usage of about 0.25MWh per year.

   In a reasonably efficient data centre, the energy use by the chillers and CRAC units is not the major wastage, so in the long term we shouldn’t focus on those units.

   Other factors start to count

   Once a data center is this efficient, other factors start to become as important as the building’s power use. Reduce your cooling still further, and the fans on your servers will ramp up.

   This might improve your PUE, because the energy they use is hidden in the IT energy sector, but to claim you are more efficient is false accounting, because those fans are less efficient than the fans in your chillers.

   Improving your PUE by turning off one set of fans and turning on another, really isn’t very honest.

   And another widely-held idea turns out to be not quite so simple: raising the temperature of the air you put in won’t always reduce your energy use.

   You might expect that inlet higher air temperatures mean less energy used, because less energy is used cooling the air down. But in fact, to cool a server with warm air, you need more of it. And that means your fans have to pull more air through, and that uses more energy.

   Or why not redesign the server, so a normal flow of warmer air will keep the processor cool enough? Because that basically means a bigger heatsink, that’s why.

   As well as using more rack space - and requiring you to build or buy more data center room -  a bigger server will use more resources to build take more energy to ship. The embedded energy costs (and quite possibly the financial cost of the chunkier servers) will outweigh any savings.

   Actual usage

   And then there’s the actual usage of servers. If you pile all your virtual machines (VMs) onto some of your servers, and turn some of the others off, it might seem like a good idea. But on the other hand, servers actually use more energy depending on how heavily they are loaded, so it might be better to spread those VMs around a bit.

   Or why not position the workload more carefully in actual space? Even with a very well-separated cold aisle, the server at the bottom of a rack will be cooler than the one at the top, if the cooling air flows up through the rack.

   If VM managers are aware of the physical layout of the server room, they could place workloads according to energy efficiency metrics.

   But wait, with all these ideas for tweaking power usage beyond the chillers and CRAC units, I’m pretty sure we aren’t actually getting to the limits of what we can do. We’re just reaching an interesting next stage. 

   -->

   Bookmark or Share this article

Login to comment.