Using fuel cells to power server racks instead of hooking data centres up to the electricity grid could potentially realise significant opex and capex savings, but would require a radical rethink of data centre design.
A group of Microsoft researchers believe that using fuel cells to power data centres could potentially result in an "over 20% reduction in costs using conservative projections", cutting infrastructure and power input costs. Fuel cells could also cut the carbon dioxide emissions by using hydrogen or natural gas to generate power.
In addition, gas grid reliability, at least in the US, eclipses the electricity grid and could eliminate the need for diesel generators and UPSes, which could instead be replaced with on-site gas tanks.
Ana Carolina Riekstin, Sean James, Aman Kansal, Jie Liu, and Eric Peterson of Microsoft Research canvassed three possibilities for using fuel cells in data centres: Either at the point where the centre is normally connected to the grid, at the server level using internal fuel cells as a replacement for conventional power supplies, or at the rack level – the option which the researchers used as the basis for their cost savings calculations.
Decreasing fuel cells' size so they could fit within servers would result in decreased efficiencies, the researchers note in their paper, No More Electrical Infrastructure: Towards Fuel Cell Powered Data Centers. 'Utility-level' – generating power from fuel cells at the point where a data centre would normally connect to the grid would be easy to implement but would entail retaining the internal power distribution infrastructure of a data centre.
Rack-level fuel cells would do away with data-centre-wide electricity distribution for servers, although this would require implementing fuel distribution lines. Having rack-level fuel cells would also mean that servers could be fed DC power directly, meaning server power supplies, which covert alternating current to direct current, could be replaced.
A great deal more study into the end-to-end reliability of fuel cell based systems and the supporting infrastructure that would be needed, such as start-up systems, is necessary before such a scheme could become a reality, the researchers note.
One challenge for the use of fuel cells is "load following": The ability of fuel cells to quickly respond to power requirements, such as servers booting up or a change in server load. Although some changes might be able to be absorbed by servers' internal power supplies, batteries, supercapacitors or "load banks" might be required in some cases where there is an instant change in power requirements. If power changes can be foreseen, fuel cell power production could be increased ahead of time.
"From our experiments with real servers, the only time that the server can cause large load change is in the startup and shutdown process," the researchers note.
"Interestingly, unpredictable events, such as rebooting a server in software or server software crashes do not cause any significant power change. This is because the electrical components are still powered in the reboot process.
"So, as long as single server spikes are handled internally, large load changing events are known to the management system. If we stagger server power on and off events over time, a single server sized battery can be shared by multiple servers in a rack. Thus the fuel cells do not need to aggressively over provisioned."
There is currently extremely limited worldwide production of fuel cells (some 25,000 per year); however, the impact of their deployment in just a fraction of data centres could have a dramatic impact on their affordability. "FCDC [fuel cell data centres] can greatly change the landscape and scale up the demand," the paper argues.
The researchers examined US Department of Energy forecasts on the per Watt cost of fuel cells by 2015 and 2020 for the paper and estimated a cost reduction based on economies of scale if they were more widely deployed.
"Based on the current US volumes, and the target of reaching only 1% of projected US data center energy consumption of 200TWh in 206, leads to a price of [US]$1.12/W," the researchers state. For their cost analysis the researchers used a "conservative range" of US$3-5 per Watt.
Capex per rack per month came in at between $250.07 (for $3/W) and $262.71 (for $5/W) – compared to capex of $313.43 for a conventional data centre setup. Opex came in at $223.51 per rack per month for an electricity grid-powered data centre; higher than the $214.06 per rack per month using proton exchange membrane fuel cells.
The researchers note that for their study they had to rely on data from fuel cell researchers – their next step is to start doing hands-on research using fuel cells to power servers.