IoT + analytics – improving data centre performance

IoT + analytics – improving data centre performance

There’s still plenty more to do when it comes to harnessing AI in the DC.

Written by Phil Alsop, Editor, DCS Europe Published Monday, 11 July 2022 10:48

Not that long ago, there was a great deal of talk about the ‘lights-out data centre’ – a fully automated facility where humans were conspicuous by their absence, hence no need for any lighting. However, there’s been a gradual realisation that, while Artificial Intelligence, Machine Learning and other digital technologies have a massive contribution to make to the business world of the future, their main contribution in many cases will be in augmenting human input, not replacing it. So, it might be more helpful to talk about the ‘lights- or lighting-optimised data centre’, recognising that humans + AI together make for the perfect data centre performance optimisation recipe!

Oh, and it might also be worth tackling another potential elephant in the room – Data Centre Infrastructure Management (DCIM). Definitely over-hyped when it first arrived on the scene, and perhaps over-sold subsequently, the early iterations of DCIM tended not to live up to users’ expectations. Nevertheless, DCIM 3.0 is very much here with us, and delivering some very real value to data centre performance optimisation programmes.

Indeed, that is now the case for AI, IoT, analytics and automation more generally within the data centre. Intelligent monitoring, management and control systems are using some combination of AI-based technologies to deliver really granular data around data centre infrastructure performance metrics. 

This allows, for example, cooling optimisation, offering the reality of significant energy savings by only running the data centre cooling when its needed and at the level required. Furthermore, by understanding how the cooling plant and other data centre infrastructure operates under various loads, temperatures and humidity conditions, it’s more than possible to extend the service life of this equipment and maybe reduce the need for upgrades and replacement parts. In other words, carrying out maintenance and service work when it’s actually required and not on some pre-specified hours or days run cycle.

Computational Fluid Dynamics (CFD) analysis, the precursor of today’s digital twin, has long been used to simulate data centre conditions to help optimise data centre design and layouts, and even to ask ‘what if’ type questions. Now, the digital twin approach has taken this ability a stage further, whereby it’s possible to carry out predictive modelling, to understand how data centre infrastructure will perform over time and under various operating parameters. This also feeds into the service and maintenance optimisation process as outlined above.

However, digital twins and associated AI solutions, such as AIOps and MLOps, can also help speed up the process of fault resolution. Ideally, they will spot a problem before it occurs or, failing that, immediately after it has occurred, providing an early warning. The (potential or actual) fault can thus be rectified rapidly, often before any users notice any performance issue. 

In future, such intelligence, obtained from the IoT sensors installed throughout the data centre, and from the processing and analysis of the data obtained, could well discover and correct faults automatically, with no human intervention required.

AIOps and MLOps can also provide anomaly detection. This means that any unusual behaviour of data centre infrastructure (whether the network, the UPS, the cooling or any other M&E component) will be flagged up, as judged against years and years of historical, normal performance characteristics. The more intelligent, granular and sophisticated the AI solution, the easier to detect even the slightest variation from normal.

AI can also be used to distribute workloads both within the data centre or, more likely, between different data centres. As edge computing takes hold, alongside the already developed cloud model, the opportunity of deciding the optimum location for applications, data and even associated IT infrastructure grows in importance. AI can help decide where applications can be run optimally, based on a risk/cost analysis process – in-house, on colocation infrastructure, in the cloud. AI can help decide when, for example, data held at an edge data centre should be transferred to a larger, more centralised facility, again based on a range of risk/cost-based criteria.

Furthermore, such AI-based selection could be extended to the data centre’s energy supply. Clearly there are many factors impacting data centre power purchase, but it’s not impossible to envisage a scenario where AI informs the data centre operator of the supply reliability and cost of a range of renewable energy options ahead of any decision.

And then we can move on to the deployment of actual robots within the data centre. Gartner predicts that, by 2025, half of all cloud data centres will deploy robots (1). And where the cloud leads, others tend to follow. 

What will they be doing? Equipment upgrades and maintenance; monitoring (avoiding the need to install sensors on equipment, which can be particularly tiresome when it comes to retrofitting IoT devices on data centre infrastructure); and physical security checks for starters.

If at this point you are feeling somewhat overwhelmed and/or frightened as to the extent that AI could take over the data centre, remember these two key things:

  1. For better or worse, as our reliance on the digital world continues to grow, so does the complexity of the IT and data centre infrastructure on which it relies – to the extent that humans are simply not able to keep up with the feeds and speeds. Or you would need so many skilled humans to keep up with real-time applications, that the cost would be prohibitive. AI and automation are vital for digital business to thrive.
  2. Despite 1., humans will still be required to set the parameters within which the AI solutions operate and/or to interpret the results of the AI solutions’ work, on which to base a whole variety of subsequent IT/data centre performance-related decision. Intelligent, intelligent decisions if you like.

 

What has all this to do with you if you are a colocation customer, or someone evaluating the colocation option? Well, you don’t need to understand too much about AI, IoT, analytics, automation, digital twins, AIOps and MLOps, DCIM and the like, but it might be a good idea to check that your chosen colocation provider does. Otherwise, you could just find yourself stuck in some ‘roadworks on a business B-road, while your competitors are disappearing down the hazard-free Digital Highway!

 

(1) https://www.gartner.com/en/newsroom/press-releases/2021-11-01-gartner-predicts-half-of-cloud-data-centers-will-deploy-robots-with-ai-capabilties-by-2025

BCMath lib not installed. RSA encryption unavailable