‘Today we are in the accelerated computing era, [which] brings new demands and new opportunities for infrastructure and system design and new technologies that are available today. We are introducing a new architecture, one that is disaggregated intelligently, so that every single element can be optimized for performance and scaled independently,’ says NetApp CEO George Kurian.
Intelligent data infrastructure technology developer NetApp Tuesday unveiled NetApp Novus, a new data architecture aimed at supporting AI infrastructure providers’ operating environments on which AI factories can maximize the efficiency and utilization of millions of GPUs.
NetApp Novus is the most important new product introduction NetApp has made in a long time, said CEO George Kurian during his Tuesday keynote presentation at NetApp Insight, held this week in Las Vegas.
“Today we are in the accelerated computing era, [which] brings new demands and new opportunities for infrastructure and system design and new technologies that are available today,” Kurian said. “We are introducing a new architecture, one that is disaggregated intelligently, so that every single element can be optimized for performance and scaled independently. One that is architected for the performance needs of the 100,000-plus GPU clusters, the unpredictable performance and concurrency requirements of the most demanding agenting workloads, and which can assimilate data across tens of thousands of systems into a single namespace to help you turn all of that data into knowledge.”
[Related: NetApp Bringing Ontap Storage Natively To Oracle Cloud Infrastructure]
NetApp Novus can help accelerate research by letting researchers run complex simulations and models in real time or biologists harness tens of thousands of computers to watch every protein molecule in a cell, thanks to its performance of 100 TB per second and ability to access over 1 Zettabyte of data with a single namespace, Kurian said.
“Welcome to NetApp Novus,” he said. “Just like the [NetApp] Data Fabric extended the value of the NetApp platform and its technological capabilities from the enterprise data center to the world’s biggest public clouds, Novus is designed to provide the architecture for the accelerated computing and AI era to help humans transform data into knowledge,” he said.
The impetus for NetApp Novus comes from the fact that while flash storage eliminated performance bottlenecks in the CPU-led world, the transition to a GPU-led world means the demands on bandwidth are much higher than what traditional storage infrastructure can offer, said Arindam Banerjee, NetApp’s chief platform and technology officer.
A single GPU sitting on high-bandwidth memory can crunch up to 8 TBs per second of data, but because the rest of the data infrastructure cannot keep up GPU utilization could fall to as low as 5 percent, Banerjee told CRN in an exclusive conversation.
“What that means is your most expensive asset that you’ve paid for in your data center is not being utilized,” he said. “And this cycle keeps on evolving. Every year we see a new GPU family, like when Nvidia announced Vera Rubin last year. That was significantly faster than Nvidia Blackwell, for example. This is a trend that keeps on evolving, and the remaining parts of the infrastructure, mainly the data part, now needs to evolve along with it.”
When building AI factories and large-scale clouds and neoclouds, the operators need exabyte-scale namespaces that can serve the massive amounts of data required for all their GPUs, but they are typically not storage experts, Banerjee said.
“Even until last year, when we talked to our customers 10,000 GPUs was state-of-the-art,” he said. “This year, people are talking 50,000, to 100,000 GPUs in a cluster. Now, if you have to serve 50,000 GPUs from the namespace and keep up a constant line rate of 2 GBs per second, that’s 100-TB-per-second bandwidth.”
Pure performance is not the only issue these providers face, Banerjee said. GPU density at the AI factory level means requirements for power consumption and data center real estate, multitenancy and management ease are also top of mind, he said.
NetApp Novus is a new architecture that federates the namespace across many data clusters by disaggregating metadata from the data because it’s very hard to scale metadata and data together on the same infrastructure or hardware, Banerjee said.
“We are taking disaggregation to the next level,” he said. “Having said that, it still works on open protocols or standards-based protocols like pNFS [parallel NFS]. Even at this level of scale, we are not going to ship you a client to install. We are going to ask our customers to operate with standards like pNFS. And even though we have new architectures here, the data is still stored in OnTap. Get all the goodness of OnTap, all the resiliency, end-to-end data integrity, security, all of those OnTap features that customers are used to.”
NetApp disaggregated data and metadata because of the difficulty of scaling the two together, Banerjee said.
“Keep in mind, metadata is small transaction I/Os, and data is large sequential I/Os,” he said. “So if they share the same hardware resources, it’s very difficult. This bimodal I/O nature is very difficult to scale. That’s why we had to separate them to scale one independent of the other. The separation provides another very important feature in that the metadata layer now can federate the data layer across many OnTap clusters.”
While NetApp Novus is aimed at massive Zettabyte-scale environments, NetApp’s channel partners have a very important role to play with the technology, said Jen Prenner, NetApp’s senior vice president of product marketing.
“As we think about where NetApp Novus will go in the future, it will be running on our A90 flash storage systems, and those are something that our channel partners sell to our customers as well,” Prenner told CRN. “So that service will definitely, over time, be something channel partners lean into. As you think about our data protection push, our new partnership with Commvault, and the AIDE [AI Data Engine], these will all be things we will train and ramp up channel partners to know about, talk to customers about, and utilize to further move NetApp up the stack.”
NetApp Novus is definitely not aimed at small shops, said Ned Engelke, CTO at Evotek, a San Diego-based solution provider and NetApp channel partner.
“We hear about scale, performance needs and the ability to control data once you’ve distributed it around geographically, and we’re really struggling to pull that together into an architecture for our customers,” Engelke told CRN. “So a technology like this has the potential of being very, very helpful for us.”
Evotek has several customers that have scale they can’t address today without a new approach, Engelke said.
“They’ve got data silos and performance constraints,” he said. “So what does Novus mean for us? My hope is that we’ll end up being able to produce different architectures for our customers that can distribute their data and get it processed and secure.”
Engelke said that the data a business uses for training AI models is often just the part of the iceberg floating above the water in relation to their total data estate.
“I have customers who have exabytes of data that they don’t know what’s inside of it or they can’t discover it,” he said. “As a result, they can’t use it, and they’re just paying to store it and regretting that expense every day. My hope is that we can start to chip away at that problem with our customers.”
Those customers include pharmaceutical companies that have accumulated 20 years or 30 years of sequencing scans or sequencing runs they have stored in the hope that they could find it again one day, or that have designed some other intellectual property around chip development and might need to discover that data to respond to a lawsuit, Engelke said.
“So what do those guys do when they can get access to all this data?” he said. “Well, they can protect themselves better. Maybe they could go back and find lessons learned from previous dead projects. …The whole point of what we’re doing at our company is trying to find areas [in which] our customers can’t do something today because that’s the only way that will earn us the right to be in their environment. Increasingly, they seek differentiation from using their data. They can’t find their data. I hope NetApp Novus helps us to facilitate and accelerate those conversations.”
