AI-managed data centre thermal optimisation – it’s nearer than you think
June 7, 2018 11:40 am
Leave your thoughts
By Dr. Stu Redshaw, Chief Technology Officer, EkkoSense
Despite best efforts, even the best run data centres still have cooling and thermal management issues. With cooling now representing around 30% of a data centre’s operating cost, it’s more important than ever for organisations to be focused on thermal optimisation.
However, for true cooling optimisation, it’s necessary for data centres to start going further and getting more granular. At EkkoSense we believe that when a data room is carefully mapped with appropriate thermal data fields, a whole new level of understanding and cooling efficiency is possible. This inevitably means more monitoring, and reporting temperature and cooling loads more actively – ideally in real-time.
With ASHRAE now suggesting as many as three temperature sensors per rack, achieving this level of sensing would typically require around 10x more sensors than are currently deployed in today’s data centres. Unfortunately, it’s still rare for data centres to sense to this level. That’s probably a key reason why, when we recently analysed over 70 major data centres, we found that 11% of racks weren’t actually ASHRAE thermally compliant. That’s a problem, because without comprehensive sensing you really can’t determine which of your business-critical racks are compliant and which aren’t.
Achieving entirely new levels of data centre thermal compliance
To address this, organisations need to work out how to build a rack-level detailed map of their data centre estate that displays all their cooling and thermal performance in real-time. And it’s only by then combining this kind of granular cooling and thermal data with smart monitoring and analysis software that organisations can start to track their data centre cooling loads in real-time – a valuable intelligence to enable thermal optimisation decisions to be made.
To achieve this kind of true thermal optimisation requires a proven, safe process that’s based on thousands of real-time sensors and expert spatial models that combine to remove the uncertainty from data centre cooling. Until recently this was a barrier due to the market cost of sensors. However, using the latest Internet of Things (IoT) enabled sensors makes this possible for less than 20% of the cost of one of the traditional cooling units. For the first time, this level of sensor deployment is accessible.
By combining this kind of sensor installation with the real-time optimisation capabilities of the latest 3D visualisation and monitoring software, you can now not only ensure ASHRAE compliance across your entire data centre estate, but also start to unlock significant data centre cooling energy savings. We have found that by adopting this kind of approach we can now deliver an impressive average energy saving of at least 23%, with a typical payback of less than 12 months.
Data centres are still spending too much on cooling
With today’s typical cooling unit utilisation rates only averaging 34%, the reality is that organisations are still spending far more than they need to on expensive data centre cooling systems.To address this data centres need to become much moreprecise in their operation, and they certainly shouldn’t be having to uniformly apply space, power, cooling and other inputs across their data rooms.
It’s only when data rooms are carefully mapped with all the appropriate data fields that these new levels of understanding and efficiency becomes possible. To do this properly we estimate that more than 1,000 sensors are required for the typical data centre, enabling the measurement of a range of previously unknown factors including energy usage, heat outputs and airflow (above and below floors) – exactly the kind of information you’ll need to evolve towards the next generation of data centre AI applications.
Delivering true cooling optimisation
Once this real-time, rack level data is collected and analysed by a 3D spatial model, specialist software can start to determine the quality of a location, identify what needs to be done to improve that quality, and even to warn operators of specific areas that are at risk.
Having access to real-time, rack-level data provides exactly the data platform needed for the kind of software-enabled real-time decision-making and scenario planning capabilities that data centres need if they’re to evolve towards true cooling optimisation – effectively removing the uncertainty from data centre cooling and ensuring that all of your racks remain ASHRAE thermally compliant.
Thanksto its unique blend of low-cost wireless Internet of Things sensors, powerful 3D visualisation software and proven thermal optimisation capabilities, EkkoSense can provide the real-time cooling data needed to manage assets more efficiently. Havinganalysed hundreds of data centres and thousands of different racks, we’ve been able to develop a proven data centre optimisation approach that means we can be 100% confident about guaranteeing the levels of cooling energy saving that we suggest to clients after carrying out one of our free site surveys.
Once our team has surveyed a data centre, we take the real-time data from our Internet of Things sensors and apply our high quality software algorithms to those findings. It’s only this combination of real-time sensors and intelligent software that allows us to build the in-depth dynamic simulations we need to effect meaningful thermal change within data centres.
This is an important step on the journey towards truly AI-managed precision data centres. This has already started with the creation of intelligent feedback loops that analyse airflow data into ‘Zone of Influence’ modules that we can then combine with standard BMS systems to enable automated zone-by-zone data centre cooling. Next will come the addition of true ‘What If?’ scenario analysis, using monitoring data to learn and predict data centre performance.
Available at a cost equivalent to less than 20% that of a traditional cooling unit, this innovative software-driven thermal optimisation approach also provides a platform for the kind of real-time decision-making and scenario planning capabilities that organisations will inevitably require as they transition towards AI-managed thermal optimisation within their data centres.
Categorised in: Uncategorized
This post was written by Dean Boyle