Technology has reached the stage where it is possible to handle the amount of data needed to support AI-based decisions about data centre operations, HDS vice-president of infrastructure solutions marketing Bob Madaio told iTWire.
Possibilities include using AI to optimise performance or cost.
The latter could involve watching how usage changes in order to manage cost without affecting performance, or applying analytics to data from power and cooling systems to see how changes affect costs and applications.
Hitachi's broader experience in IoT can be brought to bear on data centre operations. HDS already has the data ingestion and analytics capabilities needed to implement such systems, Madaio said.
Combining the 'software defined everything' trend with IoT leads to "the software defined data centre of things", he half-jokingly suggested, pointing out that software-defined elements could be treated as things along with physical elements.
It's possible to take data from hardware-defined storage, for example, but it can't be changed as quickly as its software-defined equivalent when the AI calls for action. So there are opportunities for even bigger improvements with software-defined technologies.
These automated decisions could be made on-premises or in the cloud, and could involve systems running in either environment. Moving workloads back and forth between on-premises and the cloud is challenging (especially where large amounts of data are involved), but a smart data centre could make recommendations about where specific workloads should be placed.
It probably doesn't make sense to try to handle all the operational data centrally, Madaio observed. As with some other IoT-style implementations, some edge processing is likely needed to separate anomalies from 'business as usual' data.
Deep analytics for data centres are already reasonably good, he said. The challenge is to allow AI to automatically make changes when thresholds are crossed, but "we're very close to that now".
Among the remaining challenges, "how do we give users a sense of what's actually changing so they can be confident of the (AI's) decision?"
It's likely that these capabilities will first appear in application-specific systems, such as the HDS Unified Compute Platform for SAP HANA, because it is easier to understand a specific application and its associated resources than an entire data centre.
This could happen in less than two years, Madaio predicted, with some simplistic versions that lack the ability to learn arriving sooner.
He sees these efforts as a race between different parts of the industry. Companies specialising in deep analytics and AI generally aren't looking for data centre applications, and systems management vendors will need to buy-in AI expertise to compete.
HDS's advantage is that it can bring its breadth of interests — including IoT, AI and platform expertise — to bear on the problem.
But Madaio expects other companies will catch on once they notice the opportunities.