Storage for AI Differs from AI-Powered Storage

Published on September 26, 2024, this market report was written by Jerome M. Wendt, president and lead analyst of Data Center Intelligence Group LLC (DCIG).

Storage for AI Differs from AI-powered Storage

In today’s AI-obsessed world technology providers tend to look for any excuse to attach the “AI” label on their technology. While providers may certainly do that, organizations should ensure they understand what the artificial intelligent (AI) label put on a technology means.

Not surprisingly, this trend of putting the AI label has shown up in storage. Specifically, some providers refer to their storage systems as “storage for AI” while others refer to their storage systems as “AI-powered storage.”

These 2 phrases may sound like they describe the same concept. However, they describe storage systems that may possess very different attributes. I recently caught up with Thomas Isakovich, the CEO and founder of Nimbus Data, Inc., with whom I discussed this topic.

AI-powered storage
The phrase “AI-powered storage” might imply that such a storage system represents the best choice for hosting AI workloads. While it might fit this use case, organizations should not automatically make this association.

AI powered Storage AI Chip resized
Rather, the phrase AI-powered storage communicates the storage system offers AI capabilities. This typically means in the form of ML to perform analytics. It may use these analytics in any number of ways to include:

Identifying and alerting to faulty or failing hardware components on the system.
Identifying performance hot spots or bottlenecks and potentially reconfiguring the system to resolve them.

Recommending specific hardware and/or software configurations to optimize the storage system for specific workloads.
Organizations may find these features useful if the storage system hosts AI workloads. This AI-powered storage can then help them better monitor, manage, and even optimize the storage system. However, storage for AI requires attributes beyond what an AI-powered storage system may provide.

Storage for AI
It specifically refers to storage systems optimized for hosting AI workloads. In discussing this topic with Isakovich, he found storage for AI typically possesses five characteristics.

These include:

Storage for AI Brain Resized

• High write throughput. Nimbus has found that most typical enterprise workloads tend to have a read bias. In the case of AI, it has found AI workloads to possess a heavy bias toward writes. This requires storage for AI to accommodate high write throughputs with low latency.
• Scalable. AI requires high amounts of data to operate necessitating that storage for AI scale to high storage capacity levels. This may equate to storage systems needing to scale to petabytes of capacity to handle AI workloads.
• Power-efficient. The larger in capacity that a storage system scales, the more power it may consume. Organizations should therefore account for a storage system’s power efficiency before selecting a solution. Otherwise, a storage system that consumes a lot of power will also contribute to escalating operational costs.
• Support for block and/or NFS storage protocols. To achieve the highest throughput, storage for AI must support block and/or NFS storage networking protocols. On the block storage side, he sees increased demand for IB due to the “insane” throughput numbers that IB can achieve.
• Little or no need for traditional storage system data management services. Isakovich has found that its customers that use its storage systems for AI use Nimbus’ data management services infrequently. They find features such as clones, compression, de-dupe, replication, and snapshots as “largely irrelevant” in AI environments. He himself describes storage for AI almost a throwback to the early days of AFAs when only performance mattered.

Storage for AI Differs from AI-Powered Storage

More organizations plan to introduce AI into their environments. However, as storage systems illustrate, they may do so in 2 different ways.

On one hand, they can introduce AI into the storage systems that deploy into their current IT environment. This approach should result in more resilient storage systems that become largely self-managing and self-supporting.

On the other hand, if organizations intend to deploy and utilize AI in their own applications, they will want storage for AI. Storage for AI supports features that efficiently provide high levels of performance and storage capacity. These include feature such as IB and block storage networking protocols with a reduced emphasis on traditional data management services.