Suggested Searches

Stuart Rogers: NASA in Silicon Valley Podcast

Season 1Nov 2, 2016

A conversation with Stuart Rogers, a NASA Aerospace Engineer in the Advanced Supercomputing Division at NASA’s Ames Research Center in Silicon Valley.

The cover art display for the NASA in Silicon Valley podcast.

A conversation with Stuart Rogers, a NASA Aerospace Engineer in the Advanced Supercomputing Division at NASA’s Ames Research Center in Silicon Valley.

Transcript

Matthew C. Buffington (Host): This is NASA in Silicon Valley, episode 15. Today’s guest is Stuart Rogers, a NASA Aerospace Engineer who works in the Advanced Supercomputing Division here at NASA Ames. We cover the origins of supercomputing at NASA and how technology drives exploration, especially for the journey to Mars. We also go into detail about his recent NASA Software of the Year award and how this software helps both NASA and the aeronautics industry. Here is Stuart Rogers.

[Music]

Matthew C. Buffington (Host): What brought you to Silicon Valley, what brought you to NASA in general, how did you get here?

Stu Rogers: Okay, well I’ve been at Ames since the 1980s. I got an opportunity to come here as part of a master’s program. I was a senior at the University of Colorado in Boulder, and there was an opportunity to come here as a student to work on my master’s for a two-year program.

Host:Was it to Ames, or was it just at NASA in general?

Stu Rogers: To Ames specifically, and I’d never heard of Mountain View or anything. Look it up on the map, it’s like, wow, surrounded by…

Host:A lot less snow than in Boulder.

Stu Rogers: You have to drive a lot further to go skiing.

Host:What were you working on for your master’s?

Stu Rogers: This was under the field of computational fluid dynamics.

Host:Okay, that’s quite a mouthful.

Stu Rogers: And back in the ’80s, Ames was a center for the development of computational fluid dynamics as well as Langley Research Center. And so, it was a healthy competition if you will between the centers.

Host:For somebody completely coming into it, what is computational fluid dynamics?

Stu Rogers: Computational fluid dynamics, or CFD, is essentially the science of solving the mathematical equations that govern fluid dynamics on a computer. And the whole reason you would do this is sort of the same reason you might build a wind tunnel. Imagine anything you might want to test in a wind tunnel to understand the aerodynamics or acoustics or anything like that, you could also theoretically simulate it in a computer and get perhaps even more detailed information than the wind tunnel. However, the interesting thing is that the mathematical equations, they’re non-linear partial differential equations, very difficult to solve.

Host:It sounds difficult.

Stu Rogers: So it at the time, in the ’80s, it was a growing science. And the pioneers of developing the software for CFD worked at Ames. Many of the people who developed all the algorithms were here.

Host:So, before even you set up these algorithms and you have the computer running these tests, did you have to study the actual — I’m sure there’s hands-on tests of fluid dynamics where you submerge things maybe with dye or something like that. Did you have to study through all that stuff and figure out the real deal first, or how does that work?

Stu Rogers: It’s actually an interesting split. As an undergraduate, you do a lot of study of fluid dynamics and different things, including the wind tunnel, but when you start to specialize in doing CFD for example, all the people that work in CFD generally stay behind the monitor working on the computer, whereas all the people that are doing work in the wind tunnels are the wind tunnel jockeys that are doing their work in the wind tunnels. And so the great thing about Ames is that we have such great wind tunnel facilities and we have had for such a long time is that there’s a real synergy between the two, and so the type of work we did as we were first developing CFD is, we had to validate that. We had to validate the models in the first place. And so, we developed, we worked with the wind tunnel people in a lot of instances to build a test we could then simulate in the computer and compare the answers and make sure we were on the right track.

Host:Obviously building a wind tunnel is a big endeavor. We have a few here at Ames, but imagine, building this is such a thing. If you have a computer model that can accurately depict that stuff, it’s easier. It’s a matter of running the system. Was the original intention to replace these wind tunnels? Is that what people were hoping for back in the ’80s when they were first contemplating this?

Stu Rogers: And that was really a big controversy. There was even a paper written by some people at Ames that predicted we would be able to replace wind tunnels within a certain timeframe. And while CFD has matured tremendously, and a lot of aerodynamic data that programs need are now being generated on the computer, you cannot replace wind tunnel tests. They really work together.

Host: It’s kind of like going to the back of the book for answers. I imagine you run the test just to make sure you’re on the right track, then you run it into the wind tunnel or the fluid tests to see, it checks out.

Stu Rogers: That’s true in a lot of cases.

Host:What are some cases where that doesn’t work out?

Stu Rogers: For some cases that are just, the physics are so difficult to actually simulate. For example, I currently work for the Space Launch System program, or SLS, and some of the…

Host:Which will take us to Mars.

Stu Rogers: Exactly. We’re going to take humans out of low earth orbit for the first time since the Apollo program. Some of the aerodynamics we need to know about, in particular imagine the rocket on a launch pad as it takes off, and it has to take off with some wind. And so you have essentially the flow coming from the side, and then as it takes off going past the launch pad, it transitions to the flow going past the nose. And it’s a very low speed, and understanding that, and then in turn being able to understand what’s the maximum winds that you can actually launch in, what environment are you allowed to launch in, that’s a really difficult problem from the fluid dynamics point of view. And so, a lot of those data are being generated in wind tunnels, whereas we just can’t do that in CFD. It takes too long.

Host:It’s just too complicated. And I’d imagine after you run some things, run the computer model, do it, does that help — I imagine the wind tunnels help adjust and tweak the algorithm, or at least to perfect it, things you didn’t see?

Stu Rogers: Yeah, that’s definitely true.

Host:Are there cases where you’ve had to — something comes out completely unexpected, that you’re anticipating something in the computer model, it’s running great, and then when you actually test it, you’re like, this is completely off, what’s going on here?

Stu Rogers: Absolutely. We still have surprises. Sometimes the wind tunnel data looks okay, and then you run the CFD and you compare them and you study it in-depth and you go, the wind tunnel had some interference here from the walls, or there was a shockwave being generated that reflected back. In other cases, in the CFD, we run that and find that our turbulence modeling was way off and we completely missed the physics and did the wrong answer. Which is why you continue to need both tools in order to make sure you understand the flow physics.

Host:We totally jumped in on this, because I wanted to know about your background, but this is really cool stuff. When you’re at Boulder, you’re doing master’s program, was it literally just a job you saw at Ames, and you went online and applied, or did you have contacts, you knew somebody, come on over?

Stu Rogers: Interestingly, the former Dean of Engineering at the University of Colorado had taken a job a year or two before at NASA Ames, and he developed a program to bring in graduate students from Colorado, so I was — this program was in the second year, so I was the second person to do this. And it was a two-year program where I came out here for the summer in 1983, and did a lot of reading and research, and then went back and took my master’s classes for nine months in Boulder. And then, came out to Ames for a full year with the goal of writing a dissertation, and got to work with a group that was developing a CFD software code, and worked with that group extensively, and learned a tremendous amount. And so, that kind of…

Host:You guys basically built this, you built it up from scratch. So, how spoiled are the new people who come in who’ve only been working on, who’ve had computers, these high, sophisticated computers their whole careers.

Stu Rogers: To give you an idea, when I first was taking my first computer class as an undergraduate, we still were using punch cards. You couldn’t even get on a terminal to interact directly with a computer, and so it was very much a batch mode-oriented thing. And just to get your code to compile might take a week, just because of the slow turnaround. And then even the first couple years here, we had a lot of people working in building 202A. We had six or seven people per room, and two terminals in each room. So we would have to either…

Host:Team up and figure it out.

Stu Rogers: A lot of people coming in early or stay late to be able to get access to the computer.

Host:To be able to do the work. Wow. Eventually, that team, is that what became — at NASA Ames, we have a supercomputing team. Is that the group that became everybody that works on supercomputing, or is it different?

Stu Rogers: The computer science part of it has generally been either in a different division or a different branch, however currently I am in one of the branches in the supercomputing division. We’re known NAS, or the NASA Advanced Supercomputing Center. And we have what is now the world’s 15th-fastest supercomputer, known as Pleiades, because it’s a cluster of cores of over 200,000 cores all working together as a single computer. So, as a heard your podcast talking about Pleiades, which is a cluster of stars — anyway.

Host:It’s a cluster of computers.

Stu Rogers: Yes, exactly.

Host:The best part about having a supercomputer and being able to run the stuff, it’s not good enough just to have it for the sake of having it, being able to put it into applications, whether it’s aeronautics or whether it’s science, to be able to use the supercomputer to validate and run tests to see how these things work. You mentioned the SLS, running those simulations. Are there other things you guys run? Do you do it for airplanes, do you do it for other types of spacecraft?

Stu Rogers: Yes, absolutely. And in fact, our branch, we have people supporting virtually all the mission directorates. We have a fairly large presence doing CFD for aeronautics, and that’s actually across the agency, that aeronautics is fairly busy using CFD to do a lot of the things it does. And in a more broad sense, the supercomputer is used for all kinds of things, from studying the origins of the universe to studying climate on the earth, earth climate models and ocean circulation models, that kind of thing. Kepler does a lot of its post-processing on the Pleiades supercomputer.

Host:Yeah, because it’s taking in so much data that people can’t even possibly sort through it. So, running it through the supercomputer makes sense. If you have one on hand, it’s a good thing to use.

Stu Rogers: Absolutely. And the great thing about the NAS facility is basically, we’ve grown from this desire to do CFD and to bring in computers that help us in that endeavor to actually now being the agency’s lead in supercomputing. So, the NAS facility supplies supercomputing cycles to the entire agency. And each mission directorate gets a percentage of the machine, and it’s up to the mission directorate to then send out the quotas to each principal investigator to whatever project wants it. They can apply, and then the mission directorate decides how much time each person gets. We now have of course a much bigger facility, because NASA pays for the facility through overhead. So, it doesn’t actually cost, each project doesn’t have to pay directly for its computer time.

Host:Okay, good. It’s not like that has to be implemented into their budget. It’s just like having a desk or the building, it’s a part of it.

Stu Rogers: Exactly. You get a bunch of, here’s a chunk of computing time you get, now go do good things with it.

Host:Okay, so for somebody who’s completely a layman, what’s the big difference between the computer they have at home, and what makes it super? Is it literally just the cores and processing power, or is there something different that brings it all together?

Stu Rogers: There’s a lot of similarities between the two, because in order to scale up and get faster and faster computers, we found we had to add more processors and basically process things in parallel. And in order to do that, we now have for example over 200,000 cores. The most economical way to do that is to use commodity items such as Intel Xeon chips, which are the high-end of Intel processors, but which you might find in a typical PC. But the point is that we take tens of thousands of these nodes and put them altogether with a very sophisticated networking capability, and that’s what makes it super, is the ability to literally cluster all these nodes together.

Host:To brag about you a little bit, you got some award recently? Yes, go for it.

Stu Rogers: I was recently awarded the NASA Software of the Year Award.

Host:Congratulations.

Stu Rogers: Thank you. We’re co-winners with a team from Langley. The software, this is something that’s been developed over the past 15 years or so, and there are two other co-developers that were contractors at the time in the late 1990s that we decided we wanted to write this software. And essentially, it’s a pre-processor for the CFD processing that we do, and it’s fairly esoteric in that it’s hard to describe, but we take a complex geometry with all the different parts, and we sort of break them up into different chunks and generate what we call grids, which are just a series of structurally-connected points that fill the entire volume around the vehicle, that allow us to solve the equations on that grid.

And the difficult part is that with the complex geometry, it’s hard to build these grids so that they all kind of fit together. So, there’s an approach to this called Overset CFD, where you randomly, you build a bunch of independent grids that randomly overlap each other, and that kind of relieves a lot of the constraints to building those grids. But in turn, you need to make them talk to each other. So, you find how they interconnect, you have to cut holes, and then you have to create interpolation stencils to pass data between the overlapping zones. And that all gets passed into the flow solver, which solves the equations, and gives you the answers, gives you the…

Host:Helps prep the information. Is it similar to, I’ve heard of the salesman dilemma of computing, where you have all of the places the salesman has to go to, or the post office, or whatever it is, and for the computer to run every single possibility, it just would take forever and generate a lot of heat. Is this similar to that, or am I completely off base?

Stu Rogers: It’s somewhat different than that. It’s basically that you have all these millions of points out there, and at their edges they have to talk to each other. And so, you need to search all the surrounding grids to find where they actually can talk to each other and provide interpolation stencils that allow the two zones to communicate with each other. And it gets very complex, if you can imagine, trying to put a grid around an entire aircraft, including the engine, nacelles, the pylons, wings, fuselage, tail, and all the other little parts you see hanging off an airplane. For example, when you’re landing and the aircraft goes into high-lift mode, and the flaps deploy.

Host:The flaps start curving down.

Stu Rogers: Yeah, so you have all these multiple bodies interacting, and a lot of interesting physics. And a lot of what drives the aircraft design is dependent on being able to build an efficient high-lift system and actually land the aircraft. That actually, trying to solve high-lift aerodynamics was the main genesis for wanting to write the Pegasus 5 software that we just won the award for. And that was being done in the late ’90s.

Host:Wow, so how long — has it been in process since then, of making this software and perfecting this software, all the way until 2016, you win the award? Has it been in process since the late ’90s, of pulling the software together?

Stu Rogers: That’s right. We wrote a contract to a company to write the prototype for this version. Some of the developers of previous versions of this software had a number of great ideas they wanted to try, so they wrote from scratch a new version of the code, and it automated some of the processes so well. And then we further, we took delivery of it at NASA and then continued to develop it.

Host:Tweak it, perfect it.

Stu Rogers: And fix all the bugs, and add more features, and really add some more functionality to it. But this was a collaborative effort between Boeing, NASA, and then McDonald/Douglas which by the end of the program had merged with Boeing. But we have continued to work with Boeing engineers, for example, and this software has been developed as part of their aerodynamic analysis process. And they still use it today. And that’s kind of one of the reasons it was a big deal for winning the award, was that we have heavy use of it in NASA as well as industry. Boeing has used it on every aircraft they’ve designed over the past 10 years. So, it’s shown a really good return in that part, a lot of value.

Host:They tend to say, NASA’s with you when you fly. But this is literally a thing. So, it’s not only the SLS for these rockets, for spacecraft, and for future airplanes, this is current. This is Boeing. These are normal, commercial people can come in and take advantage of the supercomputer, of the wind tunnels even, to help do their jobs.

Stu Rogers: Exactly. So NASA definitely helped develop a lot of the technology that is in use for aerospace companies developing their products. And of course it’s matured tremendously since then, but a lot of the techniques have their genesis at NASA.

Host:Cool. So anybody who’s interested in learning more about what you’re doing, about the supercomputing, where’s the easiest place to go? Just go to NASA.gov and search away?

Stu Rodgers: If you go to our division’s website at NAS.NASA.gov, there’s links to everything we do.

Host:Everything in the weeds you could possibly think of.

Stu Rogers: There’s a pretty good website on the high-end computing capability. And it talks about the Pleiades supercomputer, and everything we have in the division. And from there, you can get links to all kinds of fascinating applications that are being done on the computer.

Host:Excellent. So for anybody who has questions for Stu, want to follow up on anything, we’re on Twitter @NASAAmes, we’re using #NASASiliconValley. This is fascinating, Stu. We’ll have to have you come back, because I know we haven’t even scratched the surface.

Stu Rogers: It’s been my pleasure.

[End]