top of page

Micro-Grids for Data Centres - Could Nuclear Power Be an Option?

Micro-grids provide a viable power option in remote regions that cannot access primary grid systems, as well as providing an opportunity for technology users to maximize their energy plan by using on-site generation to meet a proportion of their energy consumption. Micro-grids can also operate like demand response systems, using utility pricing data to signal to the user when to turn ‘on’ their on-site generation resources to avoid consumption at peak times. This produces a variety of options which include energy availability, supply security, alternative fuel-mix and energy cost reductions.

For the majority of data centres, especially larger facilities and all of those in mature utility grids, the high availability (>99.9%) provided by the utility connection, in conjunction with the uptime requirements placed upon operations by the continuous computing demands for modern ICT applications, the limitations of an ‘off-grid’ micro-grid are very limiting. What is possible are such solutions as on-site generation with utility back-up, especially attractive if the source of that energy is renewable.

Touted as a feasible solution to a wide range of contemporary energy supply issues, such as energy security/supply and climate change mitigation, micro-grids are receiving increased attention as a means to power commercial enterprises, including the data center. In 2009, researchers at the Lawrence Berkeley National Laboratory in California concluded that data centers could realize high potential value, measured in terms of increased reliability and power quality, from micro-grid adoption, a conclusion that has since inspired demonstration projects at university campuses across North America.

We could question if university ICT demands, especially those big-data HPC systems where large data sets can be processed in ‘batch processing’ time-slots, are representative of the ‘always on and instantly accessible’ nature of enterprises that drive modern society – from search, social-networking, video entertainment, gaming and gambling, all the way up to banking and finance – but perhaps that is a question to be answered elsewhere. And, to be a little cynical, universities are often keen to be used as guinea-pigs (better known as early-adopters) if the new-technology comes at low price.

Of course, micro-grids that increase the carbon footprint of the national electrical power generation system are, on the face of it, a negative development. The obvious example that is already used is demand (or frequency) response contracts that can be applied to data centre emergency standby generators. In many locations such contracts (like Short Time Operating Reserve, STOR) offer financial incentives, both through a capacity reservation fee (e.g. £25k/MW/year) for ~100 hours usage per year and a payment at high rate per kWh for the utility draw avoided. However, in all grids the carbon emissions from combustion of diesel fuel-oil on site is higher per kWh of electricity generated than all carbon fuels (other than brown-coal) in power stations, even considering the transmission and distribution losses. For example, in the UK our utility generation is dominated by natural-gas which we combust in Combined Cycle Gas Turbines (CCGTs) at a thermal efficiency of ~57% and lose ~7% in station-consumer distribution, so, overall, we achieve a 50% thermal efficiency. Comparing this to the 35% of diesel standby generators shows that the net carbon footprint is increased when the STOR contract is enabled. The justification for this is that the STOR is only enabled when the grid is under pressure and having to start-up reserve plant in a power station generates more carbon than that emitted by the site generators. However, in the last several years in the UK, STOR has rarely been activated, despite repeated annual warnings that we are in danger of rolling-blackouts every winter.

But not all data centres find the STOR proposition attractive. The most commonly held view is that generator power comes with increased risks and the on-site generation is installed to protect the facility from the vagaries of the grid, not to support it. However, there is one limitation from the utilities point of view when it comes to data centres and that is the almost universal data centre feature of partial load. Most STOR arrangements ‘island’ the facility rather than run the gensets in parallel with the utility and with the critical load connected. The usual combination of partial-load and generator redundancy leads to the generators often running on-load at less than 30% capacity. This means, for example, that a data centre may have 1MW of installed gensets but when in island-mode only relieve the utility by 300kW of load.

So, what other on-site options are there that are based on a carbon-reduction basis? There is the option of running bio-fueled (preferably bio-gas for low NOX & SOX emissions) generators but the availability is lower than the utility (with genset maintenance taking 2% per year) and thereby needing a fully rated utility connection for standby. The maintenance costs are high, and the scheme can never approach the economy of utility power whilst fuel-supply and storage on site is always a difficult design and logistical problem. We can then consider on site solar-PV & wind.

It is easy to discount solar-PV to little more than a token percentage of data centre energy demand on the grounds of solar-insolation of 1kW/m2, low cell conversion efficiency of 15-20% and intermittence. Each m2 (i.e. of roof space) in southern UK will deliver little more than 600kWh/year – equivalent to <70W/m2 over 8760 hours, perhaps 1% of what is needed for a typical 5kW/cabinet data centre.

For on-site wind-power the data centre clearly needs to be in a rural location (where planning would be granted, no easy passage) and have land to spare for siting turbines. Taking the design to the obvious limit, where the data centre can claim to be 100% renewably powered by wind, produces an interesting problem regarding the utility connection. Firstly, the facility will have to have either a utility connection or a very large energy storage system, or emergency generators with diesel-fuel as the energy store, for when the wind doesn’t blow. Alternatively, it can install turbines that can generate as much energy over a full year that the data centre takes to run it and use the grid as the energy storage buffer. On-shore wind turbines can be relied upon to generate 33% of their capacity over a full seasonal year so a 2MW facility would have to install 6MW of turbines. When the wind doesn’t blow sufficiently hard to turn the turbines the facility will draw 2MW from the utility. As the wind strength increases the power draw from the utility will reduce to zero but as the wind strengthens further the turbines will feed the load and generate up to 4MW into the utility.

There will come a point (often, depending upon site location and exposure) when the wind strength exceeds the turbines safe operating speed and the array will brake to a standstill – whereupon the facility will be drawing 2MW from the utility again. So, the data centre operator must pay for an oversized 4MW utility connection to export/sell the excess energy. Although the facility will be net-zero on generation vs utility it will often be consuming utility power that has a carbon-content related to the national fuel-mix. The reader has to consider this fuel profile and reach their own conclusion.

It is probably worth mentioning that all ‘renewables’ do have an embedded carbon content due to infrastructure – such as concrete dams, submerged bio-mass in reservoirs, wind-turbine machines and foundations, solar silicon-cells and connection towers, foundations and cables, but what else ‘could’ we do regarding micro-grids for data centres that are very low carbon content? Well, in a series of lectures by Rolls Royce at the IET, a presentation on ‘small’ nuclear reactors pointed to a possible solution.

Nuclear fission reactors, such as those deployed in our nuclear submarine fleet, are ‘small’ for micro-grids (e.g. 15-20MW), highly available (like a utility connection), not intermittent, safe (we can argue that to the cows come home but morbidity rates for nuclear power are far lower than coal based power), fueled-for-life (20-25 years) and very low carbon. They make an excellent base-load generation source which will enable the maximum possible utilization of intermittent renewables such as solar, wind and tidal.

Let’s not pretend that the future is going to be easy with many choices. Recently Stephen Hawking predicted that civilization as we know it will end in 2600 due to a combination of over-population and energy shortage – and I for one am not willingly going to argue with his level of vision and intellect…

  • LinkedIn DLE Consulting
bottom of page