blobbotronic - stock.adobe.com
Nvidia plans to acquire object storage vendor SwiftStack to help its customers accelerate their artificial intelligence, high-performance computing and data analytics workloads.
The GPU vendor, based in Santa Clara, Calif., will not sell SwiftStack software but will use SwiftStack's 1space as part of its internal artificial intelligence (AI) stack. It will also enable customers to use the SwiftStack software as part of their AI stacks, according to Nvidia's head of enterprise computing, Manuvir Das.
SwiftStack and Nvidia disclosed the acquisition today. They did not reveal the purchase price, but they said it expects the deal to close with weeks.
Nvidia previously worked with SwiftStack
Nvidia worked with San Francisco-based SwiftStack for more than 18 months on tackling the data challenges associated with running AI applications at a massive scale. Nvidia found 1space particularly helpful. SwiftStack introduced 1space in 2018 to accelerate data access across public and private clouds through a single object namespace.
"Simply put, it's a way of placing the right data in the right place at the right time, so that when the GPU is busy, the data can be sent to it quickly," Das said.
Das said Nvidia customers would be able to use enterprise storage from any vendor. The SwiftStack 1space technology will form the "storage orchestration layer" that sits between the compute and the storage to properly place the data so the AI stack runs optimally, Das said.
"We are not a storage vendor. We do not intend to be a storage vendor. We're not in the business of selling storage in any form," Das said. "We work very closely with our storage partners. This acquisition is designed to further the integration between different storage technologies and the work we do for AI."
Manuvir DasHead of enterprise computing, Nvidia
Nvidia partners with storage vendors such as Pure Storage, NetApp, Dell EMC and IBM. The storage vendors integrate Nvidia GPUs into their arrays or sell the GPUs along with their storage in reference architectures.
Nvidia attracted to open source tech
Das said Nvidia found SwiftStack attractive because its software is based on open source technology. SwiftStack's eponymous object- and file-based storage and data management software is rooted in open source OpenStack Swift. Das said Nvidia plans to continue to work with the SwiftStack team to advance and optimize the technology and make it available through open source avenues.
"The SwiftStack team is part of Nvidia now," he said. "They're super talented. So, the innovation will continue to happen, and all that innovation will be upstreamed into the open source SwiftStack. It will be available to anybody."
SwiftStack laid off an undisclosed number of sales and marketing employees in late 2019, but kept the engineering and support team intact, according to president Joe Arnold. He attributed the layoffs to a shift in sales focus from classic backup and archiving to AI, machine learning and data analytics use cases.
The SwiftStack 7.0 software update that emerged late last year took aim at analytics HPC, AI and ML use cases, such as autonomous vehicle applications that feed data to GPU-based servers. SwiftStack said at the time that it had worked with customers to design clusters that could scale to handle multiple petabytes of data and support throughput in excess of 100 GB per second.
Das said Nvidia has been using SwiftStack's object storage technology as well as 1space. He said Nvidia's internal work on data science and AI applications had quickly showed the company that accelerating the computer shifts the bottleneck elsewhere, to the storage. That played a factor in Nvidia's acquisition of SwiftStack, he noted.
"We recognized a long time ago that the way to help the customers is not just to provide them a GPU or a library, but to help them create the entire stack, all the way from the GPU up to the applications themselves. If you look at Nvidia now, we spend most of our energy on the software for different kinds of AI applications," Das said.
He said Nvidia would fully support SwiftStack's customer base. SwiftStack claims it has around 125 customers. It products lineup included SwiftStack's object storage software, ProxyFS file system for integrated file and object API access, and 1space. SwiftStack's software is designed to run on commodity hardware on premises, and its 1space technology can run in the public cloud.
SwiftStack spent more than eight years expanding its software's capabilities since the company's 2011 founding. Das said Nvidia has no reason to sell the SwiftStack's proprietary software because it does not compete head-to-head with other object storage providers.
"Our philosophy here at Nvidia is we are not trying to compete with infrastructure vendors by selling some kind of a stack that competes with other peoples' stacks," Das said. "Our goal is simply to make people successful with AI. We think, if that happens, everybody wins, including Nvidia, because we believe GPUs are the best platform for AI."