Is Storage Design Undermining Your AI Efforts?

ai storage block

Posted on Friday, June 7, 2019

When it comes to artificial intelligence, storage is crucial. The machine learning algorithms that inform AI-powered decision making require huge amounts of data for instance.

As a result, many businesses rely on high performance computing (HPC) cluster to underpin their AI. This makes sense - speed is an essential element of AI operations. But it is not the only consideration when deploying future-ready storage.

Divergent architectures

HPC often may fail to deliver on performance expectations because of different approaches to data architecture. Classical HPC infrastructure is designed around the ‘bit’; AI on the other hand is more concerned with neurons.

Targeting design towards the wrong information architecture will limit performance. AI built on bit-based HPC infrastructure will work, but not as well as it could – or indeed, should.

Diverging priorities

The chances of a business being pure-AI or pure HPC are incredibly unlikely. Large organizations are likely to have elements of both as part of their operation. Which means that the choice is not bits or neurons, but how to accommodate both in a coherent, connected environment.

The answer will almost certainly involve composability. The underlying hardware will be modular, generic and commoditized. Infrastructure will instead be increasingly software controlled, allowing capacity to be assigned and configured according to operational need. This will then avoid issues of information architecture being designed for the “wrong” application.

Ultimately, composability should also help to end disputes about HPC vs AI or servers vs storage; none of these factors will matter once storage can be reconfigured on the fly. Once these balances are achieved, businesses will also be able to significantly reduce the cost of deploying new storage as they break out of the narrow use case scenarios defined by OEMs.

AI and HPC can happily coexist, but it may take some time before your infrastructure – and mindsets – are ready. 

Speak to CDS today to find out more.

Download article as a PDF - Is Storage Design Undermining Your AI Efforts?

More Articles

Black & white image of a speedometer in front of a server cabinet

What About the Other Storage Challenge: Velocity?

There are plenty of viable projects underway to increase storage capacity– but what about the issue of speed?

engineer looking at servers feeding into the cloud

GDPR Compliant? Now It's Time For CCPA

Your business has just six months to achieve California Consumer Privacy Act – or risk a sizeable fine.