As data centres are one of our most heavily relied upon resources, it is important that they operate sustainably. Matthew Underhill, Head of Infrastructure, Alfred H Knight, discusses his approach to greener data centre operations and offers his best practice advice for improving energy efficiency in the data centre space.
As I consider ways to develop a more sustainable approach to data centre operations, I believe this starts from the ground up. Ensuring that the building itself is energy efficient is the very first step during the design phase. Making sure that the building materials are sustainably sourced is also vitally important to ensuring a greener approach.
Utilising clean and renewable energy is vital to running a green data centre and reducing carbon emissions – coupling this with energy-efficient fuel cells can also make this more efficient, with increased resiliency.
Thermal containment offers improved efficiency by containing either hot or cold air – or both – into their specific areas. This improves the efficiency by offering greater control of the inlet air temperature, helping to eliminate hotspots – making the requirements more predictable. In turn, this can enable increased utilisation of free air cooling.
Free air cooling uses low outside air temperatures to help cool the data centre – and is one of the reasons that the Nordic countries are favoured for big data centre installations. Granted, this air still needs to be filtered and moisturised and this process requires much less energy over typical cooling methods.
To increase power efficiency, organisations can use AI analytics to monitor the power coming in and how it is consumed. Firstly, knowing what you are consuming is a start as it gives a target in which to aim for. Utilising AI controls not only the power, but the entire data centre. Learning from the past and predicting the future helps to reduce the amount of energy used.
Another way of operating more sustainably is to simply utilise more energy-efficient network, server and storage equipment as these require less energy initially. These also tend to be more powerful and as such, results in fewer resources required to run the same or greater workloads.
Making operations greener
As technology improves, we are no longer seeing Moors law in action – our processors are not suddenly more powerful, but they are half the size. In my role, this means I must to consider various ways to maintain the momentum of green operations. It requires a multifaceted approach looking at hardware, software and optimisation.
An example of this approach is storage. Utilising newer drives that allow for more terabytes but use less disk reduces the hardware requirement. The storage operating system also plays a key role – if the storage efficiencies can be increased release on release, we can store more on less.
Edge Computing, the much-loved buzzword of the last few years, offers a very interesting approach. Decentralising the data centre and distributing it could well be a way to improve operations. Running lower-powered systems closer to where the data is needed could greatly reduce the requirements for power and cooling in large data centres, but simply transferring data over the Internet has an impact on the environment. I have found this approach beneficial when quick results are required.
Another approach is to use the right tool for the right job, or in this case – the right processing silicon. Not every workload runs efficiently on your typical X86 CPU architecture. Putting in place the right processing hardware such as GPU, SmartNICs, or even an FPGA could help to achieve better compute results and utilise less power in the process.
Something I have been looking into recently, and another buzzword, is containers. I have been working on transitioning a very large set of applications that together make up a single application. However, these apps are running within a VM and require a couple of host servers. By migrating these applications to containers, I am not only able to achieve a higher density of containers on the same physical hosts, but I am also able to achieve application portability. Containers offer a very interesting way to shrink down your current VM estate which means you can fit more on the same number of hosts, with the added benefit of making your applications cloud-native.
Monitoring is vital, not just for the power and environment as mentioned above, but also for hosts and workloads. By improving monitoring, I can identify wasted resources – for example, identifying whether a VM assigned too much CPU or RAM? By sizing workloads appropriately, we can achieve a higher density with less.
Finally, I try not to buy more than I need. Turn off systems that are old and/or underutilised, dispose of the systems that are not energy efficient but also safely dispose of these systems. Year on year I am looking to do more with less.
Some of what I have spoken about only works at a certain scale. Hyperscalers are surprisingly energy efficient due to their size and ability to make commitments to greener data centres.
Many hyperscalers have been working towards greener operations for the last decade. Studies have shown data centres globally use 1% of the world’s energy, which has remained constant year on year, despite capacity increasing by over 500% in the last decade.
Potentially a quick win for smaller organisations is simply moving to the cloud.