Big Compute - HPC and Diversity: Life lessons from Irene Qualters

Episode Date: May 30, 2019

Gabriel Broner hosts Irene Qualters to discuss her career and the evolution of HPC. Irene, an HPC pioneer, went from being a young female engineer working with Seymour Cray to become president of Cray. She then reinvented herself to work in the pharma space and then at the National Science Foundation. She was awarded the 2018 HPCwire Readers’ Award for Outstanding Leadership in HPC.

Hello, I am Gabriel Bronner, and this is the Big Compute podcast. Today's episode is a conversation with Irene Qualters. Irene has dedicated her life to high-performance computing. She spent many years at Cray in key leadership roles, and she worked at Merck leading research information services. And more recently, Irene spent nine years in leadership roles at the National Science Foundation.

Last November, Irene received the HPC Wire Reader's Award for Outstanding Leadership in HPC. So no surprise, our guest today is Irene Qualters, and we will talk about HPC leadership. Welcome, Irene, to the Big Compute Podcast. Thanks, Gabriel. It's a pleasure to have you here, Irene. You have a unique experience having held leadership roles at an HPC vendor, a key user like Merck, and also in the National Science Foundation. And most recently, I actually am Associate Lab Director for Simulation and Computation

at Los Alamos National Laboratory. Oh, fantastic. Yep, I should have mentioned that. Last November is when you were switching roles, is that that's correct yeah so great experience coming from the an HPC vendor from also the as an HPC user at Merck and later in the National Science Foundation and now at Los Alamos so it's been a great combined experience and look forward to hearing your thoughts. So I wonder if I can take you to more of the beginning part of your years at Cray, which were probably key years in the development of high-performance computing. And if I ask you about those years, I always wonder what comes to mind so I was really fortunate when I came to Cray because at that point it was a startup and it was a rather unique

startup and I was the 79th employee to join but I was at a very early point in my career, just out of graduate school. And I had the opportunity, which was really by luck more than anything, to come to a group that had people who were really mid-career, and mid-career and very successful careers and among the 75 people or so, it really spanned almost every dimension of high performance computing that was in place. So from mechanical to chip design, to operating systems, to compilers, to mathematics, that small group of 75 people really had experience, had that breadth of experience. And so, you know, I was really motivated just because I was coming out of school and I really wanted to continue working on was such that they were starting to build and could now take on inexperienced people like myself. So that was really career forming. And I think for

those who come out of school or just starting their careers, I think really trying to situate yourself in a in a place where you're going to be exposed to many different things is can really be career forming it as it was for me as I said I was really fortunate that is great to hear those to hear. It's hard to imagine you that inexperienced, knowing you now. But it's a few years back, let's put it that way. It's great to imagine those years coming and joining Cray at that early stage. Can you tell us, throughout those years, were there any developments, any results, any things that you say that was quite interesting or that was great? Yes.

Well, as I said, that experience, you know, and some of it was personal. At that point, Seymour Cray was quite active and I don't think any longer any one person could really have the depth and breadth of understanding of a system design that he did and while he didn't like meetings we would meet there were maybe eight to ten or so of us that would meet with him periodically, and he would go toe-to-toe with each of us, whether it was the person doing the mechanical design, the cooling, the chip design, the memory choices, the compiler, and what its priorities were. And he was able to really go toe-to-toe with each of us and go as deep as was necessary. And so I think that was probably the end of that error but it it really gave me kind of an understanding of things at a detailed a holistic understanding and I think understanding how your work no matter how technically deep you are how it fits in a bitter bigger landscape I think that that that really stood out for me as a learning.

And so I had the opportunity to work on the compiler, to take some risks of my own. I was encouraged to do that, and to travel internationally to get a very deep sense of what the potential was for the work that I was doing. to appreciate that and to realize what a general, how generally important that is. And I, so I was, as I said, I wasn't pushed, but I was encouraged and I was given every opportunity. Even though I was the youngest and one of the few females, I was given really every opportunity to develop. So I think that was really important. I think moving from vectorization, the compiler was probably the first commercially successful auto-paralyzing compiler. And so that was really just a wonderful experience because the theories and the general approaches about dependency analysis which underlie the ability

to parallelize were really being developed in academia. And in order to implement a practical computer, you had to keep abreast of that work but you had to employ a lot of heuristics so melding a kind of theoretical work with Practical heuristics I really learned how to do that and that also became invaluable besides the vectorization, I think the really learned how to do that and that also became invaluable. Besides the vectorization, I think the broad understanding of parallelism then

carried over into the massively parallel systems and Cray was late to that game, but when it entered I really had the opportunity with the T3D and the T3E to a bit come from behind, but be able to be successful, and again, on an international scale. So it was a very fortunate and long period of time, over 20 years. Yeah, that's great to hear. And I have to say that the fact that the compiler did auto-parallelization enabled basically any existing Fortran program to run and take advantage of parallelism, which may have been one of the key elements for the big success of the craze at the time. Maybe not touted as such, but the reality that you can run all existing codes and take advantage of parallelism

was a fantastic thing that customers love. Is that fair? It's fair. I would also say, you know, we also had Moore's law at our back at that point. So it was possible the chip designs were shrinking and one was able to ride that curve. So if you could stay at the forefront, not only could you parallelize and boost the performance by 10 or more times, you also, even if you didn't parallelize, as long as you stayed on the edge of that curve, you could increase performance even if it didn't parallelize. And that was a very important attribute that it meant that someone could start out faster and only improve over time.

There was almost no downside. And so I would say certainly the auto vectorization was essential, but we also had other factors in the hardware design that played, were more in our favor at that time than they are today. Right. That's very, very good to hear. And it was interesting to hear you talk about Seymour Cray. I haven't met him, but I heard lots of stories about him. Seymour stories are common, especially at Cray. And the fact that he could go toe-to-toe with anybody

fits very well with the things I heard those years. Yes there's a lot of Seymour Cray stories and I can't vouch for all of them but I have my own. Yeah and they become part of the common folklore of what happens in the industry so it's very nice to understand where you were at. I was also involved with a great E3 operating system at the time. So I lived through that change that happened. And maybe you're a part of that and you said we came a bit late to it. But when you look back to the industry, that was an important change. We could say that today's clusters are an evolution of what we were doing at the time. Is that fair, Irene?

That's fair. I think parallelism and many varied ways to achieve it has been a theme really since then. And I think some of us, you and I and others, have really had a rich career in seeing the varied ways in which that can be approached. And it goes on until today. Yeah. Okay, so you spent many years at Cray. Anything more that you'd highlight about those years, the experience, perhaps something about the customers, some unique customers?

Yeah. I have to say that the relationships that I formed, not only with those I worked with at Cray, but we had very close relationships with our customers. And that's not to say that there weren't testy times and there weren't disagreements and arguments, but those relationships hold till today. And, you know, it was a really unique time. And I would even say that those relationships formed the foundation for what is the HPC community today. And I do consider it a community. I mean, the community has grown enormously, but it is truly a community. We know one another. There's a lot of trust. We can disagree and disagree strongly,

but it is most definitely a community and operates as a community. So I think that's, whether I was at Cray, whether I was at Merck, whether I was at NSF or even at Los Alamos, and still one always brought along those relationships and people would show up in your new world that were members of the community and there was an instant bond. And that bond was I I think, really formed by trying to achieve together things that just were not possible beforehand and things that we couldn't achieve them all. But we always had ambitious undertakings. And that continues today and evolves today. Maybe you just define high performance computing in that last sentence. That's quite interesting.

That's how probably we feel, as you said, a community of somewhat determined people, but everybody trying to push the boundaries. That's right. That's right. I mean, this is great to hear the time at Cray, and I thought it'd be nice to hear something I personally know less about. But from Cray, you went to Merck. And I'm curious, particular learnings about that transition from a computer vendor to an HPC user.

What was important at Merck? I mean, anything you'd like to share with us? So first of all, I would say that I had actually decided after over 20 years that I was going to make a change and a definitive change. And I had strong affinity with what I would call the application space of high-performance computing, and I had seen it in many different ways in many different fields, and that was one area I was looking at, the whole area area of application I felt that I wanted to stay in industry for sure I felt very comfortable in industry and I liked the focus in industry I I just resonated with me at that time and and I the other area I thought about going into was not the application space,

but was actually going more strongly into I.O. and data, which is interesting considering what happened and what has happened in that space. And in the end, and I looked at both, and in the end, and frankly, I took a year to do a distinguished visiting scientist role at NASA Goddard. I was invited there by a colleague who's retired now and a lab fellow, Milt Halem, who's another real role model for me. He's currently teaching at University of Maryland, Baltimore County after retiring as a NASA fellow some years back. But I knew that I didn't want to go into the government, but he asked me to come to NASA Goddard for at least six months, which I did. But in the meantime, I looked at Merck and several other opportunities in industry. And frankly, what drew me to Merck was that I had seen what had happened to the realm of physics in the 20th century.

You know, beginning with the theories, Einstein's theories, and really the rich role that computation played in the physical sciences during the 20th century, extending into the geosciences and climate models and seismic work. And I thought that the pharmaceutical industry and biology in general was poised in this century to make that transition computationally. And so I knew it was early stages, but I saw an opportunity at Merck that I could possibly have a role in that transformation of biochemistry, molecular dynamics in a transformation that was really based on its computational capabilities. And so that's really why I went. Things like imagining elimination of animal trials, could one could deepen the knowledge of

Of biological systems and chemical systems within animals so that one could do predictive simulations without actually doing the animal trials, much as what happened in computational fluid dynamics with essentially the replacement of wind tunnels with largely replaces wind tunnels with computational models Etc so that's why I went and you know, I think that that Still is at the frontiers. I think much progress has been made. There was last month the largest molecular dynamics simulation, a billion atoms, happened here at Merck, but the time scales on which that computation operates are still too small. They're on time scales of a nanosecond or even sometimes achieving a millisecond. But those timescales are not yet sufficiently long enough

to offer the kind of predictive capabilities that one will need to really advance the field. So progress is being made, but it's not there yet. But I really, I did enjoy the time at Merck immensely, and I saw the field of genetics start to arise and come into its own, and that's still progressing. So it was really rewarding in that sense. I would say what really caused all of the work that I was doing at Merck and that my group was doing was really focused so purely on the application space that I missed actually participating in the more in the computer science room and and being a part of that and so that that really caused me to turn back into areas that would bring me a little bit closer to the nature of computer science and how it's advanced in computing itself and how it's advancing yeah so itself and how it's advancing.

So I'm seeing first the reason to go, then the reason to leave, both make sense. And the common thread I'm seeing in both periods of your life where you went in early to the design of computers and see the magic progress that happened there. And you probably went in early into the pharma space and saw the beginnings of what's happening now, which will get to a revolution in medicine, right? And now you decided to leave and go more to your roots and more to where computer science was happening.

And then you went to NSF. And what I understand from your time at NSF is, which people may not know, that at NSF, you've enabled projects involving scientists from all across the world. And some of them led to breakthrough results. We tend to think about NSF as an American institution. But it's interesting to be interesting for people to hear a bit about those. So tell us a bit about NSF and about the projects that were enabled.

So one of the things that I did, NSF really focuses on support largely for the academic community. And so for me, this was a very deep change in that I was not involved in actually doing the work myself. And it took me more into the both the government side and the academic side and so I really decided that at this point in my career while I was ready to leave industry I still was felt that I wanted to intellectually contribute, and so I looked for more a role that I could be of service to the community that really I had participated in and really felt a deep loyalty to. And so I really wanted, I really deliberately decided that I wanted to switch more into a service. And I knew it would be a bit of a challenge because I'm not an academic. And even when I was in school, I knew that my career would not be as an academic and even when I was in school I knew that I my career would not

be as an academic and and so I knew that I would have to learn a lot because it wasn't my area while I interacted with academics you know I never you know, I'd never, you know, I'd not really operated strictly in that role, in that area. So, you know, one of the things that I wanted to say that is really important for those of us who over, have had these really wonderful careers that, you know, just have had opportunities that I think from a timing perspective, yes, one has to be ready, but timing is really important. And we happen to fall into a time and had sufficient skills to be able to have this ability to participate I think recognizing at this stage when our responsibilities our children may be grown we don't have such financial responsibilities that we

actually can take risks and take different risks than we could earlier in our career. And that risk can allow us to give back to that community. And I found it to be a quite really rich environment. I had to learn a lot. But it took me again to that mediation between where the technology is going and what are the frontiers of various scientific disciplines and how do those frontiers interact with one another and how do they, how do they, how can they use emergent technologies? And so, you know, really, it was quite a rich space. And I would say in terms of what new advances occurred, I didn't do any of those advances. So, you know, none of it was me. You know, it was really both trying to support researchers who were really doing those advances and support universities who were trying to do innovative approaches to computing, to supportive computing,

building a national infrastructure that could support that and international collaborations. So it was really their successes. I can't really lay any claim to them myself. But nonetheless, it was very enriching. pray to see, you know, particularly at the computing centers that NSF supports, to see researchers begin to explore, again, molecular dynamics. For example, Klaus Schulten, who's now deceased, and his team at the University of Illinois, looked at constructing a simulation of the HIV capsid, the hard shell that surrounds the virus, the HIV virus, and protects it, the HIV virus, from drugs and from other things that would attempt to rid the body of the virus. And so he constructed the first 64 million atom capsid shell, showing its structure and showing where ion channels might be present so that drugs could be targeted to penetrate

through that ion channel into the virus. And, you know, so those are real breakthroughs. Other groups were looking to, on the NSF machines as well as others were looking to detect gravitational waves, to search for evidence of gravitational waves from examination of the data coming off of interferometers, again, worldwide, and being able to match what they saw in that data with, again, simulations of black holes colliding, for example. And again, this is none of the work that I did but I played a small piece by supporting others as as they tried to do these simulations as they try to produce software and acquire systems that would allow that analysis to be done that That is very good to hear. And I think, Irene, you mentioned that you had a motivation for service, and I would say a good leader focuses on service.

And that's what you did those years at NFS. And I've heard you before talk about a few of these projects, like the Black Holes Colliding project on seismic Antarctic Project, and your passion came through. The fact that you didn't do them, you didn't do the project yourself. The fact that you were part of the team that enabled them was just wonderful to hear.

So I want to share that. Yeah, well, thank you. It's a privilege, really, to be able to help in whatever way one can and just to see those achievements up close. Yeah, very good. So, Irene, you've been a very successful leader in all these spaces. So I'm reflecting on the fact that very different stages, very different environments, you've achieved in all of them.

And I wonder, you gave a few tips here and there for new people coming along, but I wonder if there's something you'd like to share, any particular things you'd like to share for people coming along, the next generation of believers coming behind you. So a couple of things I would say, and these may sound trite but I think they're true nonetheless one of the at moments at at Cray that I can think of and groups of people that have very different perspectives, very different ideas. And so I'm really a firm believer in diversity and diversity in all its aspects. And by that I mean ways of thinking, very different perspectives, and you only get that diversity if you're bringing people in from different backgrounds, from different cultures that really can represent, you know, quite unique ways of looking at things and, frankly, multi-generational.

You need early career, mid-career, as well as more mature career people in order to get that mix of ideas that really advances either technologies or fields. areas such as quantum research, where one is trying to mix different disciplines. One is in training a new generation. And that being able to work in the environment and, frankly, being able to lead so that one is able to have different points of view, disagree, but then use those to take a field in a different direction, in a direction that no one person could do on their own. That's really the name of the game in so many areas now. And so I really, we need this next generation to be as diverse as we can make it. We need all of us. And that means, you know, both in the U.S., but it also means internationally and so I'm really a firm

believer in that and so I those are words that I would you know we have to struggle to collectively bring our different disciplines our different perspectives on the world's hardest problems and the most challenging problems. It's great to hear. Diversity is what's going to help us continue to innovate, come up with new ideas, collaborate in a good way. So it's great to hear your thoughts on that. Irene, it's a rare opportunity to be able to interview you.

So I want to thank you on behalf of our audience. Before we close, I wonder if there's something else, anything else you'd like to add. Well, I don't know that I'd like to add anything else, but I want to thank you, Gabriel, for reaching out. It's really a sign of this community that one can see someone after 10 years, after 20 years, and that bond is instantly there. So for talking with thank you it's a pleasure to reconnect with you it's been good years whenever 20 years ago but that's gonna take us a little but it's always great to connect to listen to you hear you to it two stages after that and hear your experience in the other opportunities you've had. So thanks again.

And in closing, I'd like to thank our guest, Irene Qualters, leader in HPC, for giving us the opportunity to peek into her career, her learnings, her thoughts, and help us pass a bit of Irene to the next generation of leaders. Till next time, I am Gabriel Bronner, and this is The Big Compute Podcast. Thanks, Gabriel. Thank you.