High performance computing (HPC): The risks and rewards

Data centre providers are being afforded new opportunities and challenges with the rise in demand for high performance computing (HPC). Research indicates that the market is likely to grow from $28.1bn (£19.7bn) in 2015 to $36.6bn (£25.7bn) by 2020. This rise in HPC requirements, typically the preserve of research institutes, engineers, government, military, and technology giants, has been accelerated due to the rapid increase in advanced technology being deployed in a wider variety of sectors.

Data centres: Keeping up with the demand

This increase in demand, while offering great potential to the data centre markets, brings added complications that need to be addressed. Growing consumer and enterprise technologies, such as the Internet of Things (IoT), virtual reality, and big data, are driving the predicted growth in HPC requirements. In order for data centres to keep up with the heightened demand, the increase in space dedicated to HPCs in data centres is likely to become a common occurrence.

Space is often a major consideration, especially when data centres are located in cities where square footage is often at a premium, such as in London or Manhattan, or in multi-purpose buildings where other floors are taken up by office space. Many firms see it as a more viable option to open a new site where they can have a suitable infrastructure in place to meet the needs for HPC. This will however be a significant investment so it is important to understand the current and future demand likely to come through for HPC ready data centres.

Keep it cool

In addition to space considerations, other facility upgrades are required in order to meet the demand for HPC. Data centre providers are going to great lengths in order to minimise their cooling costs, with some providers installing their data halls in cold weather climates like the Thor data centre in Iceland. Facilities are advised to run at temperatures anywhere between 20 - 30°C to achieve an optimal environment for the servers.

With the installation of high performance computers however, their processing power is significantly higher, so heat management is of vital importance. Some HPC providers have tried to ensure consistent cooling by deploying liquid cooling, larger fans or conductive cooling methods. Typically, the rate at which heat is produced is greater than the rate at which cooling accelerants or fans can dissipate this heat. It is therefore important for centres to be built with systems in place that far exceed the maximum cooling requirements, especially if there is the possibility in the future that HPC systems will be installed.

For those systems already in place and wish to install HPCs, again significant CAPEX spending will be needed to upgrade data halls to meet the required needs. Alternatively, moving to cold weather climes has enabled Verne Global’s Data Centre to be cooled naturally by outside air. By their very nature, HPCs run at much higher processing powers than regular servers. Measured in floating-point operations per second, or FLOPS, the fastest HPC as of last year ran at 33.86 quadrillion FLOPS.

Do you have the power?

Directly tied into the first two requirements for HPCs, space and cooling, adequate processing power is essential. The necessary power infrastructure must be able to cope with these demands in order to fulfil the requirements of both the HPC and the mid-range servers, depending on the data halls configuration. Those looking to install HPCs need to factor in the provision of high density power and how this has implications for the cooling requirements and the following environmental impact.

New data centre designs need to take into account the ramifications that HPCs can have on the cooling and power requirements, even if they are not currently installed on site. Those retrofitting existing data centres to accommodate HPCs need to be aware that without significant CAPEX spend, their site is unlikely to be optimal for the running of high performance computers. For those looking for data centres that have HPC capability, it is important to be aware of the space and growth potential, cooling requirements available and the necessary power provisions to provide such high processing speeds.

By having these necessary capabilities, customers can be reassured that the right technologies are in place to grow their estates. But in order for this to be successful it can’t be at the expense of the mid-range market. Focusing an entire site on HPC will considerably limit your customer base and therefore it is imperative that any investment is still able to account for traditional mission-critical applications as well as increased customer demand for dense configurations - by doing so you will significantly enhance your ability to futureproof your facility.

Greg McCulloch, CEO at Aegis Data

Image Credit: Shutterstock/Ralwel