PUE is a good data centre efficiency measure, but Peter Hopton thinks FVER might go one better
Metrics, they are not always perfect, but are a useful tool for us to measure our performance and improvement. PUE is a commonly used metric in the data centre industry, but there is room for more – and one of the most promising is called FVER.
PUE (power usage effectiveness) measures the waste in supporting mechanical and engineering (M&E) equipment by assessing the ratio of overall power to IT equipment power. Despite a little marketing abuse and users moving some of their loads into the IT equipment, PUE has been a success for the industry, driving common M&E efficiencies from wasteful to rather efficient.
FVER joins the PUE queue
Now, enter a new metric supported by the British Computer Society’s Data Centre Specialist Group (DCSG), Fixed to Variable Energy Ratio, or FVER for short. FVER is not to be ignored, it’s the brainchild of Liam Newcombe, the man who led the best practice element of the EU Code of Conduct for data centres.
FVER attempts to target waste in the whole system (software, hardware, M&E and all) like PUE targets waste in M&E. FVER assumes your data centre is made up of a sum of two loads, a fixed load that would exist if the data centre was inactive, and a variable load that would be maxed out when the data centre was full to capacity. You can use measurement of output work and power consumption to establish (and if necessary interpolate) what these loads are and FVER is then calculated:
FVER = 1 + Fixed Energy/Variable Energy
(where normally Total Energy = Variable Energy + Fixed Energy)
Like PUE, the target score is 1, but that is unlikely to ever be achieved. Real data centres will have a higher value – and less efficient data centres will have a much higher value of FVER. The concept is described in detail in a white paper from BCS: Data Centre Fixed to Variable Energy Ratio metric (DC-FVER).
Will FVER work? I hope so – the theory behind it is sound, as normally a data centre environment’s power consumption can be modelled as approximately linear, (ax+b), the sum of a fixed load (b) and a variable load (ax) that is approximately directly proportional to utilisation. By targeting the ratio you start to think about reducing the fixed load, which means an underutilised data centre will become much more efficient.
Will FVER catch fire?
To put forward an example, if your data centre’s average server utilisation is ten percent and you are given two options, reduce the fixed load or reduce the peak power consumption whilst keeping the fixed load constant, which would you choose? The answer is that reducing the fixed load will be nine times more effective than reducing the peak power consumption.
Like all metrics, it’s not going to be perfect, systems like liquid cooling for example will have a worse FVER than an air cooled system despite being much more efficient – as fans are a strong (but wasteful) variable load. FVER will however cause people to think about the fans inside their servers, encouraging them to throttle down. FVER also targets software, encouraging users to reduce server power consumption when the software is underutilised.
In summary, FVER is a simple metric to address a complex problem, and combined with measuring other metrics like PUE and a little bit of sensible behaviour; FVER could have some big benefits for the whole industry.
Peter Hopton is founder and technical director of Iceotope.
How much do you know about green tech? Try our quiz!