Simulations and Open Source Python Projects – Chris Harrold, Ansys

This is Textron TV. Hey guys. Thanks for the throw.

We're here with Chris Harrell who is the developer experience program director for answers and we're going to be talking about open source software and simulation and how all that comes together Chris. Welcome the show. Yeah.

Thanks a lot. Micah super super super excited to be here. Thanks for having me.

I don't think a lot of people have it in their heads exactly what it is that we can do today with simulation. So maybe walk us through it a little bit in terms of how the core technology works and then maybe also explain who's using this is it people who want to contribute to open source projects or people who are consuming open source software? Yeah.

Absolutely. So I mean simulation in 30 seconds right anything that you use consume ride on sleep on fly in those products. All of them are being simulated prior to manufacturing, right?

So the idea is you cut down on waste on defect you understand how that that part that widget. Entire airplane is going to react in the real world under real world conditions and the software that answers provides. Handles, all the really super complex math and physics calculations required to do that simulation.

And we do that across pretty much anything. You can think of we we can simulate it. So from the very smallest component within something all the way up to how and it has a set like an entire airplane will perform.

That's what the software does as you might expect. Not exactly a domain for the novice or the layperson to get involved in. So we we have a pretty high bar of knowledge and skill required to use the software and one of the things that we recognize is we need to open that up.

We need to make it more approachable more accessible more usable for a bigger audience. And that's really where our path is going towards open source, and this idea of allowing programmatic access to simulation, right? So when we talk about open source, I think we're following kind of the You know like the red hat model, right?

So you red hat is open source, you can use it you can consume it but they also have sort of the commercial wrapper around it where they have support and Enterprise grade feature set and and all of that. What we're doing is is very similar in that we're allowing that programmatic access that open access into the core solver engines which are our commercial, you know software that we sell right and I think it's twofold to answer your second part of your question. It's for both.

It's it's for people who want to use the solver technology in new and more sort of broad ways that that support whatever their specific goals are for programmatic simulation and then obviously there's a big open source Community out there people like to contribute especially and projects that are that are interesting or doing interesting things and we welcome that Community to come and participate in open source as well from a contribution perspective. So do you think we're at the point where we can do what if simulations so I can see how a piece of code might impact the whole build entirely and we can kind of go through that a little bit and have some understanding before we upload stuff into that building. Yeah, and I yeah, yeah, I mean we're seeing that already in a lot of cases especially I mean you think about autonomous right autonomous driving.

I'm in aerial vehicle, you know any kind of that that spacecraft right that are out millions of miles. There's no way to to change that programming on the fly. So a lot of that what if scenario programming is really part and parcel of the overall like what we do for digital Mission simulation or even just down to like specific component simulation where you have a lot of embedded code in that particular widget product spaceship, whatever it is, right?

It's I mean, that's what's that's what we need simulation to be right is just a series of what if this decision engine. What if this happens then? Oh, I this can't this part can't be here because it's going to break or it's not going to perform the way I want it to under these conditions and that whole concept, you know leads us down roads towards Ai and ml for creation of both physical and virtual digital parts.

So code and the physical manifestations of that in the real world. It takes us through, you know, the kind of volumes of data that we're going to need to get to that space. It's It's incredibly powerful and I've honestly for me as a developer, as you know, deep background in in writing code and building software and infrastructure.

It's kind of humbling to watch it all come together. But yeah, we're that what if simulation absolutely critical to this this next wave of simulation for sure. In this all this running the cloud somewhere.

I assume it takes a significant amount of horsepower. But where how do I access something like this? Yes, so so cloud is a huge part of opening our platform for Access, right?

So traditionally and and probably as you said, it takes a lot of horsepower. This won't come as a huge surprise traditionally. Our larger customers have run large compute Farms.

We've added GPU as you would expect of course for the math capabilities associated with that and the performance and speed and cloud is the next logical evolution of this. I think we're gonna see kind of two-fold right? We're going to see that You know standard AWS e Microsoft Google ecloud model, you know here where we've got customers that want to let us handle all the infrastructure.

They just want to run the simulations. They want to run right and we're building front end access for that. And then on the other hand you're going to have customers who want to keep their You know their simulation to themselves basically, but they want the cloud experience.

Right? So we're already deploying a lot of our simulation engine into kubernetes clusters so that you can do you know on demand deployment spin up spin down all the kind of stuff. I think that you're gonna see people that are gonna want it, you know, and I I said I have a really deep background and infrastructure and systems architecture.

And you know, this is the experience people want right? They don't want to just have a server farm that sits there and and gathers dust 90% of the time, you know in between simulation runs, right? So the cloud is absolutely key for us to really Embrace this whole opening and digitization of civilization across, you know, everybody who might potentially want to use it.

And that gets to the core technology because we've been talking about Cloud native for a while. But you know, it's a buzzword on one hand but on the other hand, it's also you know the cool thing about something like kubernetes is you can scale up but down and that's the part of the thing that people forget sometimes is we can do a lot of interesting things now because we don't have to necessarily stand up resources for that are dedicated to a specific application. They can be released.

Once the thing is over. Do you think people are starting to understand that because it seems to me, you know, we went through this first era where we just uploaded existing monoliths, but the question, you know to people understand how what can be done and we're about to get to some sort of New Era where things we thought not possible or possible. Yeah.

I I yes right in like I say Amy you you know as well as I might even around here long enough like Well, you know Cloud originally was oh my gosh, it's so much cheaper to put everything in the cloud. Here's my giant monolith and joy and like that's a terrible model for cloud, right? So where we're really spending time where we're really investing is is in the actual platform technology under the hood, right?

So, like I said kubernetes has been a big investment for us in terms of component sizing our solver technology to be more robust and Cloud Deployable. There's always going to be an element of monolith and I know like that that We're like, it's dirty word in Cloud, right? Oh, I have a monolith.

But I I jokingly I call them our boxes of math, right the the solvers that actually solve all the physics equations, then they're not you're not gonna break that into a microservice. Like that's just not gonna happen because the intensity of the calculations involved in solving those equations just doesn't lend itself to that and like like I said, I don't buy into the whole like oh monolith immediately equals bad. What I see is it's about the consumption of it on the front and exactly what you said.

I think yes, people are starting to see. Oh, it's not just about scale up or out right like, oh I can I can spin up many thousands of you know receivers for this and notes for this. It's also that and when I'm done they can they can go to sleep and I can I can actually reduce costs or reduce footprint in that way or shift workloads to other things right?

I mean, that's the other benefit that think people are starting to understand is like Before it was like well, my absolute high water mark was X. So I needed x amount of CPU at all times available reserved blocked for this well, but if that's only your high watermark for a little while, you know, what can you do? How can you balance that out?

What else? Can you share that with? How can you how can you shift things around?

I think the that we're starting to see that really seep into these these operations and infrastructure teams that we're working with that are running alongside of our you know, really hardcore engineering simulation experts to say like how do we really start to optimize this environment to run? And that? Yeah, I think we can't continue to operate in this whole like oh if you go to the cloud what you what you do is just copy and paste what you had on a physical server on your Cloud Server Like.

That's just that's not the right model. Anyway, I think people are starting to get it because the benefit of the cloud comes not really from the scale up and out. It comes from the scale down or the soft shut down.

Or when you're not using it, right, so yeah. Yeah, if I go back a couple of decades people were kind of getting concerned about the safety of airplanes and there was an effort by engineering teams to kind of lock them down and kind of make them more resilient in general and now we're in this era where more organizations than ever dependent upon software. So do you think that we're on the cusp of something we were here where Engineers are going to start locking down the capabilities of software so it doesn't break so often and we're not putting so many things at risk and simulation is part of that equation.

Yeah, I mean, yes, right and we already feel the effects of that. Of course. We we work with a huge number of mission critical or safety critical customers, right automotive industry Aerospace defense, right?

I mean we touch a lot of things and places that I'm sure you and I take for granted on the one hand, but on the other hand our obviously very important to be safe and I I see so much. Shift and we talk about this all the time Ajay are CEO mentioned is mentions it a frequently that you know simulation is becoming more and more important for validating that that things are going to work the way you want them to work or the way you expect them to work and I think you know the simplest example of that is like when you think about like a car crash test right like a single crash test, you know people over you wasted a car and you think about the value of the car, but what people don't think about is the the engineering and the research and the time and everything that goes into that. I mean single crash test can said, you know an automanufacturer back a million bucks and that gets spending when you have to do 20 30 40 of them.

So having the ability to simulate that stuff without having to spend that kind of capital and literal physical capital too right is becoming an increasingly important. And so to your question. Are we gonna lock down the map?

I think we've already largely done that and that's kind of goes back to what I'm saying. It's like it's always gonna be a monolith. Core there's always going to be that protected box of math.

That is. That is just right that is trustable, you know in and I talk about this. I just gave a talk a couple weeks ago about when we start getting into Ai and ml that's driven by simulation, right?

We start doing AI driven designer or machine learning assisted design the whole idea of like explainable AI or verifiable AI is going to come into play because if you can't trust the math that solved the equations right then how can you trust what the AI is telling you should should actually happen in the end result and I think you know without us protecting that core asset which you do we take very seriously, how can you trust the simulation right? If it's there's no such thing as like, yeah, it's mostly okay and simulation, right? It's either precise and accurate or it isn't and I you know, I I knew way back but I I trusted the answer simulation stuff for for what it is because we take the steps in the effort to make sure that that's a trustable, you know Bedrock of these simulation answers.

So yeah. How will simulation work its way into devops workflows? We have testing and we have quality assurance, you know, is there a thought in your mind about how and where this fits or how I might invoke it within that devops infinitely.

Yeah, I mean, I think it's gonna come into play from the literal physical infrastructure all the way up through the scaling and operation so that a day-to-day right? Like I said, one of the things that I that I'm really exciting about ansis's ability to simulate is the ability to start at the most granular smallest part of the overall, you know whole but simulate all of the pieces in between so, you know when you think about devops and I you know, I I'm an Ops guy from way back. That's I got into this career ages and age to go building data centers all over the world and you know people I think even today people kind of like, oh, I have to build a data center.

Okay, get some racks shove some servers in it. Like, you know wire them up. Okay Life's good.

But when you start thinking about the the actual like where does devops start? People, you know, we talk about serverless and all this stuff and it's like ever there's still servers there. They're still run hot.

You still need to cool them. You still need power. You still need all these capabilities and I think what this is gonna allow people to do is really start to Envision the the whole mechanism of their data center from a holistic perspective all the way from the physical beginning through to the logical operation.

And then when you when you speak about, you know, like code hardening and I want this I want to know that my my standard Baseline sort of run of operations is is X right. I'm gonna simulate that I'm gonna simulate simulate it not just in the context of the software which is important, but it isn't the total picture. I'm gonna simulate it in the context of my power capabilities my cooling the physical space that I have.

Am I going to need to add capacity? Am I going to need to you know, crank up the AC? going to need to get more power draw, you know again, we think about that high water mark and Those of us that aren't working in data centers anymore, you know working sort of like, oh, I'll just deploy 10,000 computer notes.

That's no problem. Right? Everybody's got that line around but yeah, it's like that's a that's a big deal when you start thinking about the infrastructure and everything else required to do that.

And so from a devops practitioner standpoint. Yeah, like there's a ton of stuff that simulation. I think he's gonna provide and then now that you've opened up access to it.

Now that you've made it programmatic. That's where you start to say, okay. What where how can we expand on this and we can create these libraries of essentially data center components?

Hey, you're gonna go by and a Dell or an HP or whatever server, right? It performs like this. It has this power drive this cooling characteristic.

Here's what it is at maximum load. Here's how the air flows through it all that kind of stuff and you can literally programmatically stack a data center before you ever physically build it and see not only how is it going to perform in the physical space and in the meat space, right? But also then I'm gonna deploy 10,000 compute nodes on this stack of service.

What is that look like, how does it run? What do I need to do that? It really opens up a lot of Possibility for I think for a lot of operations and devops type folks to to integrate those two mindsets really seamlessly as part of their overall workflow for managing those those devops environments.

So will we then apply AI models to the simulations to surface up recommendations to optimize for different outcomes that we're looking for? Yeah. Absolutely.

Yeah the the path to to the through the machine learning say AI assisted simulation or did we call it AI driven design. It's kind of four steps, right? So you would simulate the smallest thing that you can and the reason well not the reason but one of the big reasons that programmatic simulation this this the stuff we're doing with open source and and granting access to the underlying solvers.

It's so important is because if I set up a simulation, right so I go into the UI plug in all the variables and I hit run I go down the down the hall I get a coffee. I say hi to everybody I come back, you know some amount of time later that's simulation finishes. I get the results and then I tweet values and then I hate running again and you know, and and you repeat that maybe if you're really good you get 10 or 20 of those.

On a day right with programmatic simulation, you feed it an array of variables that you want to simulate against you hit run. You let it go in the background you do other stuff and maybe you get you know a thousand done today and a thousand done tomorrow and a thousand on the next day. Now you start to build that Corpus of information you need to train those machine learning models.

That's the part that manual simulation you can you can do it, right? Obviously. It just takes way longer than you want it or then becomes feasible, but now that we have that we can start to train those those machine learning models against the the various components.

So after you've sort of simulated that finished product right and you can come back and then you can you can wrap all that stuff up in that in that now trained AI model and before you ever. Simulate anything even as you're designing as you're laying it out, you know the AI can pop up and say Hey, you know, it looks like you're laying out a data center. You want some help with that, right?

Oh, I wouldn't put those servers next to each other. And here's why that kind of stuff. Oh, I wouldn't use that type of deployment model because it has a tendency to fail here or we've seen it break.

Here it really All you know, like I said from the from the base physical level all the way through the operational level and the software level and the code level it's going it's going to get to the point where it can pop up and point stuff out to you before you ever have to run a simulation now that's that's a future State we're building towards that obviously we've got we've got a ways to go that's not gonna happen tomorrow. It's something that will happen with adoption, which is again why open sources are important to us and why we're opening up these libraries to not just our customers but to the public right to be able to contribute and evaluate because there is a lot of innovation that's gonna have to happen to get from where we are now, it's where we all be and and the the Community Driven approach. That is what what accelerates that for us, right?

So cool exciting times indeed Chris. Thanks for being nice and sharing your As my pleasure Michael thank you so much for having me again. com as our developer portal, and we also have our developer form that links from there folks can get access start contributing or see what we're working on and we'd love to have them.

All right guys back to you in the studio.