Elnur - stock.adobe.com

Pure Storage CTO: How AI is changing storage demands

Pure Storage’s CTO stresses the need for adaptable IT systems to support AI workloads, advocating for open architectures and flexible storage platforms to avoid supplier lock-in

The growing use of artificial intelligence (AI) is driving more organisations to rethink their data storage strategies, according to Pure Storage’s chief technology officer, Rob Lee.

“With AI technologies, either here now or coming in the future, even historical data that was previously archived on the cheapest option suddenly needs to be performant, secure and modernised,” Lee told Computer Weekly during a recent interview in Singapore, adding that those requirements are necessary for organisations to extract value from enterprise data.

In designing IT systems, Lee said organisations will need to ensure the systems are flexible enough to keep pace with the breakneck speed of AI developments.

“You can’t predict the future like with traditional business applications,” he said. “Your best strategy is to maintain maximum flexibility to adapt to industry changes.” This contrasts with the traditional approach of over-provisioning for anticipated needs, a practice now challenged by the unpredictable nature of AI workloads.

While cloud computing offers a degree of flexibility, Lee distinguishes between capacity flexibility and technological adaptability, stressing the importance of avoiding technological lock-in. “Making the wrong technical choice and needing to re-platform because of insufficient performance or lack of application support is costly,” he warned.

Pure Storage’s “as-a-service” model provides consumption flexibility, but Lee highlighted the greater value of combining that with flexibility in technology choices. This will require the use of open architectures and open networking, allowing for integration and future adaptability.

“With every choice, you should be thinking about how much it locks you into a technology choice and how much flexibility you have to adapt,” said Lee, pointing to Ethernet as a prime example of a ubiquitous open architecture.

Read more about storage

  • Originally driven by Intel’s now-defunct Optane storage class memory, Parallelstore offers massive parallel file storage targeted at artificial intelligence training use cases on Google Cloud.
  • A major storage upgrade is enabling the Auckland War Memorial Museum to safeguard its vast collection, ensure digital sovereignty and enhance visitor experiences using AI.
  • From data storage to intelligent data infrastructure – that’s the plan from NetApp, which has announced data curation for AI as well as additions to its ASA and FAS storage arrays.
  • We talk to Charlie Boyle of Nvidia about AI impacts on storage I/O, key performance metrics for storage for AI, and storage attributes required across differing AI workloads.

But the dominance of certain suppliers in the AI ecosystem, particularly chipmakers, can pose challenges in maintaining open architectures. Lee acknowledged this, noting that “there’s a lot of learning to be done to understand what’s tightly coupled to a specific stack and what’s not”. He believes the market will eventually prioritise balanced performance across a range of AI applications over optimisation for specific types of workloads.

Lee also discussed Pure Storage’s expansion of its FlashBlade portfolio, particularly the introduction of the E family, designed to replace traditional hard disk drives. “This has been very well-received globally, especially in the Asia-Pacific region,” he said, highlighting the appeal of achieving flash performance at disk costs.

Furthermore, the ability to combine the E family with higher-performance FlashBlade offerings provides the benefits of traditional tiering without the complexity and performance drawbacks, added Lee. “Because of the uniformity of our hardware and software architecture, we can apply high-performance compute power to data residing on more cost-effective storage without moving it around,” he said.

On the aftermath of Broadcom’s acquisition of VMware, Lee acknowledged it has prompted some customers to re-evaluate their virtualisation plans. “It’s definitely a topic of concern,” he said.

While some organisations are exploring alternatives to VMware, others are adopting a wait-and-see approach. Pure Storage aims to support customers regardless of their chosen path, whether they choose to use managed VMware services from the hyperscalers or Kubernetes-orchestrated options, said Lee.

On support for Nutanix, which some VMware customers have migrated to, he remained tight-lipped. “We work closely with a lot of vendors in the space, and over time, we look forward to being able to work with Nutanix to support our storage offerings,” said Lee. “We don’t have anything at the moment, but stay tuned.”

Read more on AI and storage