The Diesel Question?
Author Ian Bitterlin, Portman Associate, CEng FIET MCIBSE ATD
My title pays homage to the book written in 1865 by William Stanley Jevons where the question was coal, not diesel, and the consuming industry was the cotton mills of Manchester (nicknamed Cottonopolis at the time), and not data centers. That book included Jevons’ Paradox, which still has application in today’s internet regarding broadband capacity and resulting data growth – but that is another topic for another day.
For the past few years, and increasing recently, discussions about diesel generators in data centers have centred around the fuel itself and the resulting C02 emissions. It has been interesting to note that little has been discussed about particulates and gases in the exhaust, rather concentrating on carbon with a climate change agenda, but I will address this here.
There is no doubt that data centers want to be seen to be green and avoid the publicity such as that Google received from Greenpeace about paying for very low-cost brown-coal (lignite) generated electricity in Oregon, which happens to release the highest rate of CO2/kWhe at >1,100g. It is worth noting that currently Germany, home to the EU’s largest data center estate, is dramatically increasing burning indigenous lignite to substitute for the nuclear generation it has turned off and to the reduce the dependence upon Russian gas. Unlike Google, a single data centre in Oregon that had a 25-year PPA at $2.5c/kWh, Germany has the highest electrical grid load in the Europe, so the carbon impact is far greater. If the reader is interested further in grid carbon, from the last hour, month, year etc, then app.electricitymaps.com is a good resource.
The point to be made is that the carbon footprint of the utility affects our view of the impact of using diesel as an emergency fuel.
Despite the language of ‘net zero’ there is, for the foreseeable future, no such thing as ‘zero carbon power generation’. The reason is simply that the infrastructure (buildings, foundations etc) and fuel mining/extraction, processing, delivery, and installation etc must be amortised over the life of the generation and power distribution plant. This include concrete, steel, copper etc. For example, a wind-turbine may have a useful service life of 15 years and the carbon-emissions at the surface of the blade is clearly zero – but the foundations, mast, blades, generator, variable-speed electronics, step-up transformer, and cables connecting it into the grid have a high carbon footprint that must be accounted for over its lifetime. A high proportion of the infrastructure cannot be recycled.
So, we first need to understand what we mean by ‘low carbon’, although it isn’t catchy or sensational enough as ‘zero-carbon’ as used for public consumption.
The carbon content of electrical grid per kWh is well defined by Intergovernmental Panel on Climate Change (IPCC) 2014. The IPCC is the United Nations body for assessing the science related to climate change. There are other localised rating systems, such as EU-ETS and ENTSO-E 2021, but here we are going to look at the principles, not the detail, nor try to pick the rating system that best suits a certain agenda. For example, IPCC rates coal as 820g steam-coal, not the more polluting 1,152g brown-coal etc.
It should now be clear to the reader that burning diesel-oil in a data center can range from a ‘carbon reduction’ impact in a grid that is dominated by coal-fired generation, to a ‘carbon increase’ in a grid dominated by nuclear and/or river-hydro. The absolute value of the reduction or increase will depend upon the hours run per year for the data center generators and the fuel that is used to feed them.
Diesel-oil is a convenient fuel for any emergency power generation system. The fuel is non-flammable with a high flash-point of 700⁰C and has a high Net Calorific Value (NVC) per litre, hence it is safe to store and space saving, especially if you want storage >12 hours. Diesel engines are also very simple in that they are compression ignition machines and do not have an electrical ignition system. A diesel genset can start and accept load in under 10s (it is even an NPFA code requirement in the USA) so the ability to start quickly and reliably is also a key feature.
For many years, suppliers of natural-gas fuelled generators have proposed them as a ‘cleaner’ solution to diesel, and some publicity suggests they are ‘green’ in some way, but the three features of space, flammability, and ignition complexity are critical when a data centre application is considered. Natural-gas, methane, takes up considerable storage volume and is highly flammable. Gas gensets are excellent for long-time operation with a grid gas supply and if storage is not an issue for the application, then turbines, rather than reciprocating engines, and particularly combined cycle, would be preferable for fuel consumption although starting time and black-start arrangements place design limitations on the data center UPS and cooling systems.
Fuel storage & consumption
Fuel storage (and therefore consumption) is clearly an issue for the designer and different fuels have varying issues of flammability and volume. From 2020 onwards an issue that had negative publicity is the consumption of diesel fuel, and actions were taken (and publicised) to replace diesel with Fatty Acid Methyl Ester (FAME), Hydrotreated Vegetable Oil (HVO), Natural Gas (Methane), or Hydrogen.
Below we can compare the fuels:
- Diesel oil 950g CO2/kWh 42MJ/kg, 1,360kg/m3, safe to store
- HVO/FAME 100g CO2/kWh Drop-in for diesel, safe to store
- Methane 520g CO2/kWh Flammable, 0.72kg/m3 , 120MJ/kg NCV
- Hydrogen is highly flammable, very light and low calorific value per kg
- 0.09kg/m3 (8x the storage volume of natural gas, 13,000x that of diesel)
- 50MJ/kg NCV
- Loses 15% of its thermal value to compress to 200bar, or 40% to liquify (to reduce the storage m3)
- The generator will require a larger capacity prime mover or liquefied fuel
However, with hydrogen we need to differentiate between Blue Hydrogen, which is produced from Natural Gas, and therefore has a higher carbon content than natural-gas and Green Hydrogen, which is produced from renewable energy.
In 2022 the only commercial Hydrogen available in bulk was Blue.
- >700g CO2/kWh for liquefied, using wind power
- Compared with coal-fired utility power at 820g CO2/kWh
The process for Green Hydrogen is energy intensive, as it uses six-eight times the amount of energy to produce than Blue, and the carbon footprint will be in the order of:
- >500g CO2/kWh using wind-power (or similar for nuclear)
- >575g CO2/kWh for 200bar, using wind power
It is important to note that the delivery distance can increase the carbon footprint. Over 25 years ago a pair of engineers from EdF wrote a white-paper about a diesel fuelled 40,000L road tanker taking 200bar Hydrogen on a 350km delivery distance. By calculation the truck engine would have consumed the same calorific value as that contained in the Hydrogen load. Hydrogen needs to be produced close to the point of consumption.
Diesel generator operational hours per year
However, it is critically important to get the scale of the consumption of diesel (or any other fuel) into focus, especially as the diesel generators are there to act in an ‘emergency’.
Clearly there are many locations in the world where the electrical utility ‘fails’ regularly but the places where data center of large-scale and high-density are built are in developed locations near to power, connectivity, and users. In these mature grids failures lasting longer than three seconds are often only annual at worst. In normal designs the generators are set for a delay-to-start of three-five seconds. During that time the UPS supports the ICT load for continuous service and the cooling system stops, waiting for the generator power to come into service about 10 seconds later. Starting ‘in anger’ is therefore a tiny part of the 8,760h in one year.
In addition to the grid failure duty the generators must be tested to ensure reliability when needed and, if the preventive maintenance and testing is done correctly (no more than monthly with 30% load) the generators in mature locations will see a total annual duty of 8h/year and never at full load.
The impact of alternative fuels on the CUE
In ISO 30134 we can find the Carbon metric for data centers, Carbon Usage Effectiveness (CUE), which numerates the carbon emissions of the energy system in each kWh consumed in the ICT load. Users usually ignore the carbon in the diesel fuel (and we may show why below) and in any utility water used in an evaporative or adiabatic low-energy cooling system, which is also a very low contribution compared to the usual utility.
For the comparison we took the national utility carbon content recorded on 16 Jan 2023 at 23:00 UK, for a variety of data center hubs (using electicitymaps.com) and the impact of 8h/year running at various fuels is shown as a percentage of the utility carbon emissions.
The date and time of the data is irrelevant as the numbers change continuously. The only purpose is to show the percentage impact of emergency generation.
For the comparison we have used a simple utility and generator load of 1MW at full load. So, for example, if you had a 10MW utility connection for your multi-tenant data centre in Europe then that could indicate an ICT full load of <6MW, due to the utility and generator being rated for the peak summer load at 100% ICT load. Partial load (i.e., not 100% design load) is endemic in data centres and so the actual load could be from 1MW when the facility is new, to 4MW at a level of maturity. As diesel engine fuel efficiency is more-or-less linear with load between 30% and 100% you can use the table below for ‘Year 1’ and multiply the CO2 by 4 for ‘Year N’.
We can now see that the impact on the overall carbon emissions of diesel fuel is tiny, in fact much less than one year for a single city-centre diesel bus, per MW of data center utility demand.
A simple drop-in of HVO (or FAME) drastically reduces (by ~90%) the annual diesel tonnage. It needs to be checked that the engine sets do not require any derating – but partial load and ‘hottest day of the year’ rating will always swamp that requirement. An additional consideration might involve continuous fuel-polishing (water and bacterial sludge removal) if not already fitted to the diesel tanks.
The application of natural gas sets (a change of engine) does not save enough carbon to compensate for the downsides of storage, flammability and starting reliability.
Blue Hydrogen is higher than Methane, so is not shown, whilst bulk Green Hydrogen is not available yet but the delivery, flammability, and storage (it would have to be 200bar or liquified) may override the advantage.
It appears that The Diesel Question, if one exists, is simply answered by going vegetarian, and we are kidding ourselves that claiming to be worried about diesel generators is going to make any meaningful difference to our industry’s ‘greenness’.
There still exists the problem of city-centre pollution from particulates and NOx/SOx, even if only for 8h/year. Whilst data centers are scrupulously clean compared to traditional cars and trucks it would be prudent to include scrubbers on the exhaust system to remove (>99%) those emissions harmful to local air quality and health, or simply bar data centers with generators in metropolitan areas.
Such a ban is not entirely unrelated to downtown Tokyo (no diesel, gas turbines instead) or parts of the San Francisco Bay Area where diesel fuelled gensets are banned. This raises the topic of fuel-cells, especially those tried by eBay in California, of the PEM/natural gas configuration. They are dependent on pure hydrogen to be clean and the use of natural gas poisoned the fuel-stacks and created hazardous waste every few months. There has been no publicity surrounding the multi-MW installation for several years.
I am not yet convinced that 100% hydrogen fuelled reciprocating generators will be reliable without a significant change to the engine design. I know tests have been carried out but have not seen much data being published on ratings. It will be interesting to see progress.
We need to focus on decarbonising the grid (unless you don’t believe in climate change, and many still doubt) before we start tinkering at the edge of data center technology, apparently with the sole purpose of greenwashing. If you do believe that climate change is real then the idea that we need to have ICT ‘always on’ will be challenged as climate change accelerates.
There may come a time, although probably not in time to impact climate change, when the ICT load can shut down fast enough so as not to lose data and not need emergency generation at all – but that assumes that the user will accept outages of service. This is probably manageable for social networking and streaming, but unlikely to be acceptable for finance and banking.
About the Author
Ian is a Chartered Engineer with more than 27 years’ experience in datacenter power and cooling following 25 years’ in rotating electrical machines and systems.
Formerly CTO for Emerson Network Power Systems in EMEA, Ian is now principal consultant at Critical Facilities Consulting and a Visiting Professor to the University of Leeds, School of Mechanical Engineering. Having filled senior posts in major UPS and datacenter OEMs, Anton Piller, Liebert, Emerson Network Power, Active Power & Chloride, he has been instrumental in key product innovations and continues to consult on new datacenter design and M&E products.
An active blogger, trainer and author of many technical papers on critical power and cooling, Ian continues to give presentations and keynote speeches around the world. A Fellow & Member of Council of the IET, Member of CIBSE & BCS, Ian sits on datacenter committees and international standards bodies, is ex-Technical Chair of the Data Centre Council of techUK, ex-Chair of the DCSG of the BCS, Accredited Tier Designer of the Uptime Institute and ex-Chair of the TWG in EMEA for The Green Grid. Ian is a Technical Expert & Expert Witness for datacenter related project disputes with HKA Ltd.
Ian has been awarded ‘Outstanding Contribution to the Data Centre Industry’ twice, by Data Centres Europe in 2009 and by DataCenterDynamics EMEA in 2015.