Artificial Intelligence enters datacenter monitoring.
Predictability
It has been a while since we last gave an update on 4D Cool, the real-time 3D datacenter heat map. Together with our launching customer we have been working hard on this first prototype and developed new tools that further enhanced the predictability of the datacenter. Predictability is the holy grail of an advanced monitoring and control system: you really want to be able to predict the effects of changes in your DC. Change is the only constant in your DC and 4D Cool helps to understand what happens if you add or change hardware. Alternatively you might want to be prepared for severe changes in your weather conditions.
Artificial Intelligence
4D Cool uses Artificial Intelligence (AI) to improve the quality of its algorithm. AI has been in de centre of attention for quite a while now. In essence there are three kinds of applied technologies: Machine Learning , Deep Learning and ‘Digital Twins’.
Many AI solutions are based on Machine Learning. In this case, the underlaying algorithm is based on the relationships found in sensor-data. The down side of this model is that it only recognizes situations that are present in the input data. No options to virtualize a non-existing situation.
Deep Learning, the other side of AI, is not based on a man-made algorithm. It creates its own just by analyzing (big) data. Although this may result in algorithms that are new (and sometimes surprising) it is also prune to unwanted results because of a biased input. Think of the Google experiment of hiring staff by AI, based on thus far successful applicants. The (unwanted) result was that AI selected only male applicants as the majority of the workforce was male.
Digital Twins is a combination of data, algorithms and a physical model. In this case it is possible to generate what-if scenarios in the Digital Twin which then predicts the outcome in the physical twin. 4D Cool has been designed as a Digital Twin solution. (more info on Digital Twin here)
What if…
The 4D Cool model, is based on computational fluid dynamics (CFD) together with a full Digital Physical model of the datacenter which improves itself by adding data from a wide array of environmental sensors in the data rooms. Hence the full digital twin of the reality. The virtual
and real datacenter merge as one which allows us to generate ‘what if’ scenarios.
You can change the humidity of the outside air, or kill an UPS and see what the effects are in your virtual datacenter without affecting the real one. This is the ultimate tool for anyone who is in charge of a complex physical entity like a datacenter.
There are so many variables in a datacenter which are all interconnected, that it has become impossible for a single person to oversee all the effects of changes in the DC. AI can help managing these changes but only Digital Twins allow for actual ‘what if’ analysis.
Next phase
4D Cool had just migrated from idea to concept when we last mentioned it in a post. It has now moved on the next stage: pre-production. Currently there are two datacenters that are using 4D Cool in their daily operations. Although there are some more developments to be done, both datacenters are very pleased with the results of 4D Cool. They both have been able to improve their energy efficiency as well as their workflows. The power of predictability assures that maintenance takes place at the right place, at the right time. Unplanned outages have decreased dramatically. (Unspecified) cost savings up to 10.000 euro per month were mentioned by the datacenter managers involved.
