Picture the scene. You’re sat in an aircraft taxiing your way towards the runway preparing to jet off on your holiday. But before you can reach for your book or lose yourself in a film, a freak snow storm descends, sending your plane back to the gate to wait on standby until the poor weather has passed. Not only is this frustrating for you and your family, it can be hugely costly for the airline in question, with every movement of the plane costing thousands of euros.
But High Performance Computing (HPC) is being used to predict weather patterns more accurately and rapidly; making aircraft take-offs much more cost effective and safe. In the second part of this series (read about how HPC is being used to redefine ship designs here), I explore how supercomputing is being used in the Netherlands to generate more accurate predictions of weather conditions in near real-time.
Amsterdam Airport Schiphol is one of the world’s busiest international airports, with nearly 60m passengers passing through its terminal each year. One plane takes off or lands every minute, equating to around half a million airplane movements each year. With up to €60,000 being burned with every aircraft movement, it’s imperative that the planes can take off and land as efficiently as possible. But bad weather can throw this process wildly off course. Being able to predict the forecast therefore, is imperative to ensuring the airport can run as smoothly as possible during different weather conditions. For instance, by optimising aircraft movements and preparing planes for take-off the minute the rainstorm ends.
This is where HPC plays a pivotal role. Due to the masses of information needing to be processed for each forecast, computational simulations have been used to predict weather patterns for a number of years.
On a non-HPC server however, generating a 48-hour forecast would typically take 38 hours to complete, meaning virtually all of the value in the data would be lost by the time it was delivered. HPC is being used in the Netherlands to generate 48-hour forecasts in less than 90 minutes, which can be delivered to Schiphol every six hours. Speed is crucial here to ensure accurate forecasts can be made on time, every time.
Increasing the efficiency of HPC
As this example shows, HPC is already playing an integral role in streamlining major operations across multiple industries. R&D teams must now focus their efforts on reducing supercomputers’ Power Usage Effectiveness (PUE) to ensure these powerful machines can be used on a wider scale. When diving deeper into the energy usage, traditionally built supercomputers have a PUE of about 1.6: for each kWh used to process data, with an additional 600 Wh required for cooling. To lower the total energy costs, scientists are building modern supercomputers using Direct Liquid Cooling (DLC) methodology, which lowers the PUE to about 1.1 and therefore uses much less energy to run. Bringing this figure down will have a tremendous impact on the use of HPC in not only meteorology, but also other industries such as oil and gas and engineering; and it’s therefore crucial that efforts are focused on this area in the short term.