Doing The Math (or "HUE, PUE, Barney McGrew...") - by Peter Judge
A few weeks ago, I praised PUE, the leading
metric for data centre efficiency, but said it was a "blunt instrument". As readers of this blog will already know,
there is plenty of work going on to sharpen it up.
There's some specially clear thinking coming from Deutsche Bank, form the look of this interview in News@Cisco, with Andrew Stokes, chief infrastructure architect at the bank,
Stokes believes like anyone involved in green data centers, that it is all about measurement. "You don't realize how much room you have for improvement- how much capacity you actually have in your data center - until you really measure what's going on," he told News@Cisco.
DB has a series
of efficiency measures. It starts with PUE, but use its reciprocal instead:
DCIE, or Data Center Infrastructure Efficiency, as defined by the Green Grid is the energy used by the IT equipment
divided by the total energy.
But how efficient is the IT equipment? Stokes has an appealing way of doing that, which is simple enough to get your head around, and should be easy to apply to other data centers. It is spelt out very well in a more detailed post on Cisco's Data Center blog,
You can measure how good your IT equipment is by measuring the amount of processing it can do (the number of instructions or operations) per Watt of energy. How much computing power are you getting per Watt?
Unlike DCIE or PUE this is not a simple ratio, because Watts and MIPs are different units. The only way to get a simple percentage, is to compare with the best possible figure you might expect. This means that DB's "Relative Hardware Power Efficiency" (HPEr) is based on a comparison with current best practice - how efficient is your server compared with what the industry can offer?
That comes from an Ideas International idea called Relative Performance Estimate 2 (RPE2).
Because best practice is a moving target, Stokes works out his own figures each year, based on the equipment available. I'd hope this figure is used carefully, as setting targets too aggressively might weight the scales in favour of replacement, despite the environmental cost of disposing of old hardware.
The final layer of the onion (or at least the deepest Stokes goes at the moment) is "Hardware Utilization Efficiency". That's just the CPU usage of the servers.
So DCIE times RHPE gives an idea of how efficiently you are generating CPU cycles, HUE shows how many of those are used or wasted. Including these factors would help shift the emphasis to using data centers better, instead of just building them well
So we have three metrics - and the beauty of DB's system is they are all ratios, and can be multiplied together. Further metrics could be multiplied in, if other layers of efficiency become important.
However, the names HUE and PUE have reminded me of the firemen/s roll call, "Hugh, Pugh, Barney McGrew, Cuthrbert, Dibble and Grub", from the vintage BBC children's show Trumpton (though TV fans dispute whether it was in fact "Pugh, Pugh...").
I suppose it would be too much to hope for a few more metrics named after "Barney McGrew", "Cuthbert" "Dibble" or "Grub"...