Intelligent Data Centres Issue 06 | Page 23

INDUSTRY INTELLIGENCE POWERED BY THE DCA Call to leading vendors We are absolutely certain that if server manufacturers work with the partners at the BTDC-1 research project, we can help them (and the entire data centre world) to slash average cooling PUEs from 1.5 to 1.004 system. At EcoCooling, we have taken all of that temperature information into the cooling system’s process controllers (without needing any extra hardware). Normally, processing the cooling systems are separate with inefficient time-lags and wasted energy. We have made them close- coupled and able to react to load changes in milliseconds rather than minutes. As a result, we now have BTDC-1 ‘Pod 1’ operating with a PUE of not 1.8, not 1.05, but 1.03. The BTDC-1 project has demonstrated a robust repeatable strategy for reducing the energy cost of cooling a 100kW data centre from £80,000 to a tiny £3,000. This represents a saving of £77,000 a year for a typical 100kW data centre. Now consider the cost and environmental implication of this on the hundreds of new data centres anticipated to be rolled out to support 5G and ‘edge’ deployment. generations of servers will use far less energy when not busy. So instead of 75% quiescent energy, we expect to see this fall to 25%. This means the cooling system must continue to deliver 1.003 pPUE at very low loads. (It does.) Also, BTDC-1, Pod 1 isn’t just sitting idly drawing power – our colleagues from the project are using it to emulate a complete Smart City (including the massive processing load of driverless cars). At BTDC-1, we have three research pods. Pod 2 is empty – waiting for one or more of the mainstream server manufacturers to step up to the ‘global data centre efficiency’ plate and get involved. As a sneak peek of what’s to come in future project news, Pod 3 (ASIC) is now, using the same principles outlined in this article, achieving a PUE of 1.004. We are absolutely certain that if server manufacturers work with the partners at the BTDC-1 research project, we can help them (and the entire data centre world) to slash average cooling PUEs from 1.5 to 1.004. The processing load varies wildly – with massive loads during the commuter traffic ‘rush hours’ in the weekday mornings and the afternoons. The opportunity for EcoCooling to work with RISE (Swedish institute of computer science) and German research institute Fraunhofer has allowed us to provide independent analysis and validation of what can be achieved using direct fresh air cooling. And then (comparatively) almost no activity in the middle of the night. So, we can expect many DCs (and particularly the new breed of ‘dark’ edge DCs) to have wildly varying power and cooling load requirements. The initial results are incredibly promising and considering we are only halfway through the project we are excited to see what additional efficiencies can be achieved. ◊ Planning for the future – Automatically adjusting to changing loads An integrated and dynamic approach to DC management is going to be essential as data centre energy-use patterns change. What do I mean? Well, most current- generation data centres (and indeed the servers within them) present a fairly constant energy load. That is because the typical server’s energy use only reduces from 100% when it is flat-out to 75% when it’s doing nothing. At BTDC-1, we are also designing for two upcoming changes which are going to massively alter the way data centres need to operate. Firstly, the next www.intelligentdatacentres.com Issue 06 23