Dell Technologies Interview: Dell’s Jay Boisseau on Data Growth Outpacing Moore’s Law, the HPC-AI Divide-Convergence and Beating Zoom Fatigue

Print Friendly, PDF & Email

[SPONSORED CONTENT]  Jay Boisseau, HPC & AI Technology Strategist at Dell Technologies and organizer of the Dell Technologies HPC Community meetings, is one of the big personalities of HPC, someone who brings energy and insight to any gathering he’s a part of. He’ll debate HPC issues with anyone, all in a good-hearted, good-humored way. In part this makes him a natural to lead the HPC Community meetings that, during the pandemic, have been held online weekly and are attended by hundreds of Dell customers, business and technology partners, Dell executives and by anyone interested in HPC.

In this insideHPC interview conducted on behalf of Dell, we talk with Boisseau about his career path, including serving as director of the Texas Advanced Computing Center, about big shifts and future trends in HPC and about the phenomenon, where HPC and AI intersect, of data growth rates outpacing Moore’s Law. He also disc uses his strategy for infusing life into the HPC Community meetings in an era of “Zoom fatigue,” including drawing inspiration from The Art of Gathering by Priya Parker, who, as Boisseau points out, probably never dreamed of the Zoom era.

Doug Black: We’re talking with Jay Boisseau, he’s Dell’s HPC and AI Technology Strategist. And he also runs the Dell Technologies HPC Community, a gathering place for Dell HPC and AI customers. Jay, great to be with you again.

Jay Boisseau: Thanks for having me looking forward to talking.

Black: Okay, so tell us a little bit about the community. You know, when was it established? How many members do you have? And what kinds of activities do you have?

Boisseau: Sure, well it’s the Dell HPC community, and I want to emphasize that it’s not just for Dell customers, it is for Dell staff, it is for technology partners throughout the HPC technology ecosystem. And Dell is a very proud partnership-oriented company – and we have a lot of partners in that ecosystem — and for customers, including users, developers, IT administrators, C-level execs, various roles within that community. So we have all three of those stakeholder groups, Dell partners and customers / users as part of the community.

The size of the community is difficult to characterize because in the Zoom era we’ve gone online, and we have a weekly event now every week instead of just the three meetings per year. The meetings typically draw a few hundred people, and now the online events are drawing 100-plus minimum and sometimes 200-plus people on a weekly basis to these online events. And again, the online events represent all three of those stakeholder communities, and we make them open to everybody. So in fact, we have people that aren’t yet partners and aren’t yet customers because we’re really trying to educate everybody about the importance of HPC and how we can advance it and how we can use it better in more ways.

Black: I attend a lot of those meetings. I know you’ve got one on Arm and HPC that I’m really interested in. But what are some of the topics that have drawn especially large audiences?

Boisseau: The one we have coming up that you just mentioned, it’s going to draw a lot of people because there’s a lot of interest in whether Arm is ready for mainstream HPC. We certainly see Arm in everything, from embedded devices to smartphones to laptops, we certainly see it in the number one HPC system, Fugaku. But in between there still a lot of gaps in terms of market presence. But the technology is very interesting. And in the HPC Community, we not only showcase the products and solutions that we make, the products and technologies that our partners make and the techniques and applications that our customers and users develop, but we do sometimes venture into the space of technologies that maybe we want to evaluate for future use.

And so I fully expect this upcoming event with Brent Gorda of Arm and my good friend Jeff Kirk of the (Dell) CTIO office will draw a lot of people. And we’re going to do a fun format, it’s going to be a “Shark Tank” format where Brent pitches that Arm is ready for mainstream HPC and Jeff asks him the judge-type questions about that. So that’ll be big.

But past topics that have been really well attended? Let’s see, whenever we have Dell thought leaders from the CTIO office present, as we did just recently, we have a huge audience. When we have topics of great interests, like on prem versus cloud, or hybrid cloud or the cloud space for HPC, that’s been big. When we have certain key vendor partners present — Intel, AMD, Nvidia — great partners of ours, when they present updates on their latest technologies those draw very large audiences as well.

When we have key partners, such as the Texas Advanced Computing Center (TACC), which I used to lead, and other great Dell customers, when they present, they draw large audiences as well. I think the main thing I’d want to say is we really never have a small audience. It’s been amazing to me in the Zoom era that so many people tune in week after week.

Black: So Jay, everybody meets on Zoom — I know you guys use Zoom — but what makes the (HPC) Community compelling and the members coming back week after week in an era of Zoom fatigue?

Boisseau: Doug, this is one of the favorite questions that I get asked because it’s so important. We’ve all accepted that Zoom and technologies like it are here to stay, and they’re powerful and they’re useful and whatnot. But there are differences. Meeting in person gives you an energy, there’s the tactile experience of shaking the hand, there’s 3D views of people, It is a different and in many ways better experience. So how do you create an environment on Zoom that is also engaging, not just informative, not just like watching a TV show, or something, but is actually engaging?

We take that really seriously in the HPC community, and so we’ve tried to figure out what the Zoom equivalents are of some of the wonderful things about meeting in person. So we encourage everybody in Zoom chat to introduce themselves to each other in the same way as if they were around the coffee and donuts at a conference. They can’t exchange a physical business card in the Zoom community, but they can type in who they are, their role, what they do, and they can add their email link or LinkedIn link in there, and people connect to each other through the Zoom chat during these meetings.

We also try to keep it a little bit more informal and fun. And we, we present great information from thought leaders. But we never want it to feel overly processed like a professional video, we want it to be professional people sharing content in a more intimate, informal way, in a community. So we encourage a little bit of informality while at the same time assuring our speakers we’re going to have an incredibly talented and intelligent audience on board, they should share as if they’re giving an extended TEDx presentation, a thought leader presentation. Even if they’re talking about a technology or a product or solution, their job is not sales or marketing, it’s education, it’s information, it’s inspiration to a bunch of like-minded people that want to hear that content.

We also not only change the topic every week and the speaker, but we’ve even changed the format to try to use different formats, to keep it fresh and engaging and leverage that principle of neuroscience, that people love familiarity, i.e., week after week, but they crave newness by a new topic, a new speaker, new format.

So these are all really important principles, and I learned some of them from the great book, The Art of Gathering by Priya Parker, who probably never predicted the Zoom era, but you can even apply principles in that book to the art of gathering online. You have to honor your attendees and respect them, and realize it’s real time they’re investing and that it should be worthwhile time. And you as the host, you’re responsible for honoring that. Me and Janet (Morss) and Karl (Cain), we take our role very seriously in hosting the HPC Community.

Black: And you bring a lot of life to those meetings, and they’re generally fast paced, entertaining and good energy. So tell us a little bit about your background. You mentioned TACC, the Texas Advanced Computing Center, where you were director until 2014. So I suppose you’re moving over to Dell was kind of a natural transition?

Boisseau: Well, in some ways, yeah. The first part of my career was as a PhD in astronomy, and so I was a scientist that used supercomputers. And then the second phase of my career was being in that middle role of not using them or making them but integrating them and offering them and supporting them for that science community. And that was incredibly rewarding. And that’s an important task, because you know, on one end you’ve got technology makers making technologies and servers and storage networks. On the other end you’ve got scientists using these things to change the world and advance human knowledge. But there’s a key middle ground that great centers, like TAC and SDSC and others, fill around the world to help researchers use them.

But it is interesting to now also be at a technology provider, a global technology giant like Dell, that makes everything from laptops to petaflop-scale supercomputers and literally everything in between. And understanding how the sausage is made, and how we can try to provide better solutions for customers like TACC and end user customers that provide their own on prem capabilities, to enable their researchers, their developers, their engineers to solve the really challenging problems that they want to solve. So yeah, it is, it is interesting to be at a great technology company like Dell and understand how to make the technologies and all the decisions that go into that process.

Black: I like that your title combines HPC and AI, and I’m, in my own small way, I’m trying to use that hyphenated, two-word, phrase as commonly as I can, because they’re getting so tied at the hip. Now, there’s divergenciess between the two as well, we’re seeing AI supercomputers, etc.

Boisseau: Well I have thoughts on that, Doug. I believe I sold out by using that in my title, because I’m the first person that, if you ask – ‘Are HPC and AI the same thing?’ – I’m the first person to say no. And they’ll say, ‘What do you mean?’ And I will say HPC is a solutions approach, HPC is how you build high performance computers. So computing systems that are more powerful than regular computing systems – semantically, that’s what the name means, high performance versus mainstream.

And Doug, you and I know that a few decades ago, (HPC) wasn’t clusters of commodity technologies, it was proprietary hardware architectures with proprietary software ecosystems. And so the word ‘supercomputer’ was invoked because it was either a computer or a supercomputer, and there was nothing in between. So you didn’t have to define it, you know it when he saw it.

Now commodity technologies have made it possible to get tremendously better price/performance through clustering. And companies like Dell have embraced leadership because of that possibility. And so HPC is a little bit more of a vogue term than supercomputing and it’s a different methodology. It’s clustering these technologies together, these mainstream technologies, but now you cluster them with networks and software that make it look like a virtual supercomputer in many ways. And that is a fascinating time to be in this space, because of that.

AI is a workload, and AI has a whole set of workloads, and some of them run on supercomputers and HPC systems, and some of them run on $10 sensor devices that you have around your home. So AI is really a set of workloads, HPC is a technology approach. There are traditional HPC workloads, there are AI workloads that run on HPC. But I sort of feel like I sold out a little bit by merging a solution type and a workload type in my title. But I don’t apologize for it because it’s so important to recognize the connection between large scale AI workloads and the need for HPC solutions to execute them.

Black: What I like about it is that we’re talking about a certain class of AI workloads, if you will.

Boisseau: Right, exactly.

Black: Staying at that 30,000 foot level, what do you see as a few – one, two or three – of the big trends shaping HPC, HPC-AI, over the next 12 to 36 months?

Boisseau: Well, you know the old phrase that’s attributed to both Niels Bohr and Yogi Berra: predictions are hard, especially about the future. But that doesn’t stop any of us from wanting to make them.

The easy one is the growing adoption of AI and the increase in successful use cases for it. So we see AI used in academia, in enterprise and government. I think what you’re going to see is a lot of return on investment successes, especially in enterprise, where everything is measured in ROI and TCO and ‘If it makes sense to do this.’ All these companies are getting familiarity with object classification and computer vision, and natural language processing and recommendation engines and anomaly detection and these other workloads that are fundamental AI workloads.

And they’re starting to get good at it. They’re understanding the solutions we’re making and how to use them. And of course, we’re a part of that, we not only make solutions, we help our customers understand how to use them. So in the enterprise space, we’re starting to see lots more successes, and I think that is a hockey stick uptick and that’s what we’re about to see for the next couple of years ahead – lot more things moving from pilot projects and POC projects to actual deployed sustainable stuff.

And of course, we all see it in our daily lives, we use recommendation engines from Amazon and Netflix to figure out what to buy or what to watch. We all get that occasional credit card query of “Was this really you?” – that was AI used in potential fraud detection. But I think we’re going to see a lot more of that in our daily lives, and certainly our customers are transitioning much more from POC and pilot projects into enterprise use cases.

We’re also in the academic and research environment starting to see a lot more usage of AI in in concert with simulation, AI that helps guide simulation, AI that complements it in ways that frankly mean you can use less computing power to explore the right parts of the parameter space and really get to insight and discovery faster.

Of course, one of the things that’s driving this intersection of AI and HPC is the fact that data growth rates are growing faster than Moore’s Law. So we’re seeing people have access to data at scales that outpace the scale of even the tremendously fast scale of processor and accelerator development. And so what does that mean? It means that if you have an AI or data analytics problem that just barely fits on one server today, then tomorrow, it’s likely to need an accelerated server or two servers. And next week, it’s like to literally need two accelerated servers or maybe a handful of accelerated servers.

So as long as the data available to people who are trying to solve data-driven problems is increasing faster than the capabilities of individual systems you’re going to need HPC to help them execute the data analytics techniques, often now AI techniques, to get the maximum value out of that data, to extract the maximum insight.

We all know the simple analogy, it takes two points to use linear regression and draw a straight line, but it takes 10,000 or more images to train an AI model to recognize a cat. It’s vastly more computational, but it’s vastly more powerful too. And so all this data and all this increased access to data is driving more and more AI to use HPC. And I think that’s another big trend, we’ll see. So: more successful use cases transferring out of POCs and pilots into real successes with ROI and improving TCO; the increased use of HPC systems in AI, especially on the training side of it.

And then if I can offer just a couple of my favorite applications of this, I think we’re going to see a lot more interesting things happen in smart places – smart hospitals, smart factories, smart cities — as we start to collect larger amounts of data about our environment and are able to make better decisions about both human-guided decisions and autonomous decisions. It’s not one or the other, it’s both. And so I’m super excited to see how the smart infrastructure, smart places, smart cities applications grow in the coming two, three years.

Black: Okay, great stuff, Jay. Thanks so much for your time. Today wee’ve been with Jay Boisseau of Dell Technologies. Thanks so much.

Boisseau: Thank you.