Published on 15th January 2015 by Ian Bitterlin
The best example of diversity in electrical system design is the one closest to everybody – in the place where we live. Take the UK as an example; the standard domestic electrical connection is a fused 100A 230V single-phase supply that could supply 23kVA. My house is quite middle class (although we are working class folks) and has the standard 13A ring-main for the power etc.
I went around the house counting the lights, power sockets and major electrical items. The place is quite old so has small windows and thick walls so there are more lamps that you would get in a modern house but the result of my 'audit' came to 4kW of lighting (c15A but getting less as we fit low-energy bulbs).
The dominant consumers were the 13A sockets, of which I was amazed to discover a total of 33 – that would amount to 429A if they were all delivering full amps. Then there is the major items of fixed consumers – the oven, the hot-tub and the summer hot-water cylinder – a total of 110A.
So the total system load ‘could’ be get to 554A when the fuse would rupture very fast indeed. Clearly it is not (thankfully, given the tariff) and the fuse has never been known to blow. Interestingly the diversity factor is 100/554 x100% = 18%.
The data center - a diversity-free zone
So what do we do in a data centre design? We start with the full load possible at each cabinet, ignore any diversity whatsoever and multiply it up with power and cooling system losses to arrive at a design load for the utility supply. And what happens in most cases? We end up with partial load and worry about partial load efficiency and PUE etc,. Now 18% diversity may be rather optimistic for an enterprise or collocation data centre but I suspect that 30-40% would not.
While we have the exemplars of Google et al which can rack+stack and get the power density it plans for because its utilisation is nearer 60-70% than the 5-10% the rest of the world struggles with. We have all heard a version of the tale of the IT department that said that a new 10kW cabinet was being installed and 4kW of demand turns up – its nearly up there with ‘I’ll love you in the morning’.
The risk of getting it wrong
So I wonder if we will ever get to the point of truly scalable infrastructure based on realistic demands. Or is the risk of getting it wrong far too high a price to pay compared to the extra investment up front? I suspect it’s a similar situation to energy effectiveness – everyone talks about but the three key drivers for enterprise data centre design is still reliability, reliability and reliability, and strictly in that order.
Ian Bitterlin, Consulting Engineer at Critical facilities Consulting Ltd