Ending the industry’s reliance on generators

The UK government recently stated that data centres are to be classified as critical national infrastructure alongside those like the NHS and power grid. This means they will get additional government support in the case of a major incident. To improve resiliency, data centres have often relied upon generators. However, this does not mean they are necessary for every data centre build. In fact, many data centres can reach industry reliability and availability standards without a single generator on campus.  

As stated by Uptime Institute, an unbiased advisory whose mission is to report on the technical requirements for resilience in data centres: On-site power production should be considered the primary power source. Local utility power should be seen as an economical alternative only.  

Because of this and the desire from major data centre developers for a standardised reference design, we’ve seen many operators believe that they need full generator backup for every project. If, instead, they were to consider sites on a case-by-case basis, they would realise they could hit their resiliency benchmarks while improving their bottom line and sustainability targets for some developments.

The industry has largely set itself the target of net zero carbon. While many changes are happening across the full scope of data centre design, an important issue that is being overlooked is the need for generators. It’s time to reevaluate the status quo of data centre design, and that means asking: Do we need generators for our data centre?

Utility supplies and data centres

Data centre operators target a standard reliability metric, ‘5 9s’, which refers to an uptime of 99.999%; this is calculated by looking at the average time it takes equipment to fail (Mean Time to Failure) and the average time it takes to repair (Mean Time to Repair). 

Data centre campuses have grown from 10MW IT to 100MW IT to 1000MW IT and above scales. Particularly with machine learning and artificial intelligence development, these sites are becoming connected to the grid at higher voltage levels. This significantly changes the reliability of the utility networks supplying these campuses. Typical distribution network connections operate between 11-33kV and are usually fed from the same substation; as such, when you get a utility fault, it would usually take out both your A and B utility supplies. This should not be mistaken with transmission networks, which instead operate at a much higher voltage of 132kV or above. Additionally, they have a much higher level of resilience with multiple supply routes, which allows the transmission grid operator to carry out remote switching and resupply sites via alternative routes.

To understand reliability, developers must look at their data. When this information is available from the utility, data centre developers can evaluate the predicted system’s reliability by reviewing the historical performance of upstream networks and substations. This includes the mean time between failures (MTBF) and mean time to repair (MTTR) and inputting this real-life data into their data centre reliability calculations. In reviewing the electrical system holistically, you are better able to determine if generator backup capacity is even needed.

The Energy Networks Association (ENA) engineering recommendation P2 Security of Supply recognises the increased resilience of connecting to the network at higher voltages. We are currently seeing proposals for the large data centre campuses that would fall into Class E, referring to connections over 300MW and up to 1500MW capacity. Under the ENA P2 guidelines, the first circuit outage should be restored immediately, and a second circuit outage should be restored immediately to two-thirds of the capacity. It should be noted that immediately is defined as within 60 seconds to allow for switching of the grid. 

Progression through reduction  

For progression to occur and for our industry to meet the commitments of the COP21 Paris Agreement, we must adopt a reductionist mentality. First comes dismantling the belief that generators are necessary for every data centre site. The end-tenants must lead this change as it is their leasing agreements which enforce full generator backup.

Changing the perception of these tenants by showing them the data will be the first step to achieving change in the industry however, for this the utility companies must be more transparent with their network resilience as well.

A way of solving this is for end-customers to increase the scope of their site selection process, which includes reviewing the grid capacity and adjacency of major transmission substations. This is where the higher reliability of the connections can be provided, as well as reviewing the historical reliability data for these substations. Considering the grid first gives operators the opportunity to make more strategic decisions about location. They can choose data centre locations close to major substations, which enables resiliency to be provided through the grid rather than backup generators.

After this, the industry can take a broader look at what resiliency means. At this point, operators reviewing a site for data centre construction rarely look beyond resilience at a site level. Taking a step back to look at their data centre network as a whole means they can see where several discrete sites connect at different parts of the grid. Through this lens, they can provide redundancy through their IT infrastructure and enable generator backup only for vital network traffic. This would mean that if one site goes down, the network traffic can then be diverted to other data centres connected to either discrete parts of the grid or separate grids entirely. They still achieve resiliency and remove their reliance on generators alone.

Introducing generators to a site costs the operator financially, the designer in complexity, and the environment the embodied carbon stemming from the production and emissions during routine testing. In my own experience, I have worked on data centre sites in which over 50 generators have never been switched on beyond routine testing and maintenance. This is incredibly inefficient. At the same time, some data centre operators have smartly selected sites with highly resilient utility connections that mitigate the need for generators by achieving ‘5 9s’ availability without them.

We have seen real-life data dispel the myth that generators are a necessity. Yet, the industry still chooses to mandate full generator backup throughout the leasing agreements. Despite being able to find other sites better suited to their projects and reaping the benefits of cost savings and better sustainability metrics, belief in generators is being maintained. It is a fact that generators will still be a must for a lot of projects, however, it is only when operators look at their projects individually that they determine the best solution for them. It’s time to break the industry’s perceived reliance on generators and adopt a better mindset.