As a member of NetApp United, I had the opportunity to sit in on a briefing from Mike McNamara about NetApp‘s recently announced AI offering, the snappily named “NetApp ONTAP AI”. I thought I’d provide a brief overview here and share some thoughts.
So what is NetApp ONTAP AI? It’s a “proven” architecture delivered via NetApp’s channel partners. It’s comprised of compute, storage and networking. Storage is delivered over NFS. The idea is that you can start small and scale out as required.
- NVIDIA GPU Cloud Deep Learning Stack
- NetApp ONTAP 9
- Trident, dynamic storage provisioner
- Single point of contact support
- Proven support model
[image courtesy of NetApp]
Thoughts and Further Reading
I’ve written about NetApp’s Edge to Core to Cloud story before, and this offering certainly builds on the work they’ve done with big data and machine learning solutions. Artificial Intelligence (AI) and Machine Learning (ML) solutions are like big data from five years ago, or public cloud. You can’t go to any industry event, or take a briefing from an infrastructure vendor, without hearing all about how they’re delivering solutions focused on AI. What you do with the gear once you’ve bought one of these spectacularly ugly boxes is up to you, obviously, and I don’t want to get in to whether some of these solutions are really “AI” or not (hint: they’re usually not). While the vendors are gushing breathlessly about how AI will conquer the world, if you tone down the hyperbole a bit, there’re still some fascinating problems being solved with these kinds of solutions.
I don’t think that every business, right now, will benefit from an AI strategy. As much as the vendors would like to have you buy one of everything, these kinds of solutions are very good at doing particular tasks, most of which are probably not in your core remit. That’s not to say that you won’t benefit in the very near future from some of the research and development being done in this area. And it’s for this reason that I think architectures like this one, and those from NetApp’s competitors, are contributing something significant to the ongoing advancement of these fields.
I also like that this is delivered via channel partners. It indicates, at least at first glance, that AI-focused solutions aren’t simply something you can slap a SKU on and sells 100s of. Partners generally have a better breadth of experience across the various hardware, software and services elements and their respective constraints, and will often be in a better position to spend time understanding the problem at hand rather than treating everything as the same problem with one solution. There’s also less chance that the partner’s sales people will have performance accelerators tied to selling one particular line of products. This can be useful when trying to solve problems that are spread across multiple disciplines and business units.
The folks at NVIDIA have made a lot of noise in the AI / ML marketplace lately, and with good reason. They know how to put together blazingly fast systems. I’ll be interested to see how this architecture goes in the marketplace, and whether customers are primarily from the NetApp side of the fence, from the NVIDIA side, or perhaps both. You can grab a copy of the solution brief here, and there’s an AI white paper you can download from here. The real meat and potatoes though, is the reference architecture document itself, which you can find here.