Modern Industrialist Podcast


Episode 1: ”Decoding Industrial IoT: Exploring On-Premises vs. Cloud Solutions in the Era of Machine Learning”


Guest: Patrick Turley, Head of Engineering at TXI


The Modern Industrialist Podcast is for technology-focused manufacturing and logistics leaders looking to gain a competitive edge with Industry 4.0 transformation. Join Jason Hehman of TXI as he brings together experts from companies blazing the path for the Titanium Economy revolution. Topics range from advice to success stories, use cases, solutions, and more. 


On this episode of the Modern Industrialist Podcast, Patrick Turley, Head of Engineering at TXI, joins Jason Hehman to discuss the barriers to wide-scale Industry 4.0 adoption and on-prem versus cloud in the industrial manufacturing space.
About the host Jason Hehman
Jason is the Vertical Lead for Industrial IoT and a Client Partner at TXI, a company that discovers, builds, and delivers digital solutions from concept to transformational impact. With over 20 years of experience in the industry, he has certified brand strategist credentials and a proven track record of helping clients solve complex business problems through technology. 
In Jason's current role, he works closely with clients in the industrial sector to help them leverage IoT to achieve growth and unlock new value. He applies expertise in product innovation, digital transformation, and smart manufacturing to guide them through every step of the process, from ideation to execution. He is passionate about understanding consumer behavior and using that insight to impact business strategy, as well as creating products that enhance user experience and satisfaction.


Interested in connecting with Jason? Reach out on LinkedIn
Summary of this Episode:
00:00 - Introduction to the Modern Industrialist Podcast, discussing trends and challenges in Industry 4.0, with hosts Jason Hehman and Patrick Turley.


00:35 - Patrick Turley introduces himself and explains why he's called "Turley," discussing his background in tech consulting and focus on the industrial sector.


01:15 - The hosts explain the goal of sharing behind-the-scenes discussions and introduce the topic of on-premises vs. cloud solutions in industrial IoT, inspired by a workshop conversation.


02:01 - Mention of a previous workshop about on-premises vs. cloud solutions and the influence of Oxide's perspective on rethinking cloud computing models.


03:32 - The discussion turns to implications for machine learning in the context of on-premises and cloud solutions, and the considerations for industrial players.


04:02 - The importance of positioning key aspects before diving into details, acknowledging the shift in computational demands due to machine learning's rise.


06:03 - Recognizing the evolution of cost dynamics, with on-premises becoming more appealing due to changing resource needs and computational demands.


09:31 - Identifying bursty applications like model training and real-time data processing as areas where cloud or on-premises decisions have significant impact.


12:22 - Diverse computational profiles of bursty tasks vs. steady-state operations and the importance of aligning solutions with specific patterns.


15:22 - Discussing the value of hybrid approaches and reevaluating cloud solutions for industrial partners, focusing on reliability and data transfer costs.


19:22 - Emphasizing the need for balanced decisions that challenge assumptions and align with performance, security, and efficiency requirements.


20:02 - Recognizing the dynamic landscape that requires continuous assessment and curiosity, especially as machine learning gains importance in the industrial sector.


21:36 - Conclusion of the podcast episode, encouraging curiosity and exploration in understanding the complex choices between on-premises and cloud solutions in industrial transformation.
________________


Transcript:






Modern Industrialist EP 1 


SUMMARY KEYWORDS:
cloud, on-prem, industry 4.0, IIoT, machine learning, computational ai, aws, artificial intelligence


00:00
This is the Modern Industrialist Podcast accelerating transformation in the Industry 4.0 era. Welcome to the first episode in our series. Our goal with this podcast is going to be to help spur adoption of technologies that are critical to industrial innovation by talking about the current trends and challenges that we're seeing in this space. My name is Jason Hehman. I'm the vertical lead for Industry 4.0, and IoT at TXI. It's a lot of acronyms. Joining me today is our TXI head of engineering, Patrick Turley. Truly, please introduce yourself and explain to people why we call you Turley.


00:35
Patrick truly is a full name that my parents gave me. That's right. I've worked in a lot of places with a lot of Patrick's and somehow you got to differentiate yourself. So I took on the moniker of just early a long time ago. And so pretty much everybody calls me that I've been at a TXI for almost seven years coming up soon. And I've been in the tech consulting space, pretty much my whole professional career. I have done a whole bunch of different types of technology in different spaces. But most recently, a lot of stuff in the industrial space.


01:15
That's great. I think that's helpful context for bidding, you and I end up having a lot of these conversations around trends and things that we see are happening. We're talking about the challenges that our clients are facing. And I think we thought that this would be a great opportunity for us to share some of these behind the scenes conversations that you and I have been having with a broader audience. And that's really where we started with the topic that we're going to cover today. I think you and I were preparing for a workshop two or three weeks ago, and we started getting into the conversation around balancing the needs and trade offs of on prem versus cloud, how our partners in the industrial space are thinking about that. And you had a really interesting perspective based on a conversation that you'd had that kind of spurred us off on a bit of a tangent.


02:01
Um, yes, for all of our fans out there the giant sea of fans of this podcast, if you Yeah, if you haven't checked out that workshop, it's available on TXIdigital.com. Go have a look at that. It was an excellent talk with Jason and Andrew, Principal Engineer here. But a lot of that is one of the questions we were talking about on prem versus cloud in the sort of industrial IOT space. And that sort of came from a conversation that I had with the CTO at a company called oxide. These guys are trying to sort of rethink rack computing, making it a little bit more available and accessible to folks who are, you know, questioning the cloud computing promises. And in particular, I think the costing model that historically we talked about, they're trying to challenge that and, and still produce an excellent product. And they're doing a great job. By the way, the other conversation I had with him is they are, they produce a really awesome podcast as well that I love oxide and friends, very similar style thing where they just have conversations about what's what's going on in their heads, they they grab, for the most part, they're engineers together and just have conversations like this. It's super cool. You can get that on any of the platforms, you're interested in whatever podcast wherever you get this podcast, and that's what they say on the podcast.


03:32
So I think what's neat, or what I was hoping we could take the conversation today is layering on the considerations in the industrial space of, again, cloud versus on prem. But when you start to think about the implications for machine learning, and what that is going to mean more and more for players in the industrial space. How does that consideration that decision point between those two options start to take a different level of importance, or what other considerations start to come into play?


04:02
Yeah, um, I mean, I think the big, let's position some things first, before we sort of dip into that I was gonna take five steps back and a good run at one of these. But if I were to say, like, historically, why have we thought about cloud versus on prem? I think one of the big things is, you know, cost, it's very expensive to run servers these days. It's very cheap to have things in the cloud. We're gonna get into that here in a minute. But I think the move towards machine learning and all the AI presence, I think it changes the computational needs that we have for the business of most of these places. So if you think about like a, like a factory or like a smart factory, as compared to somebody who's, you know, they've got their administrative applications running. Those are Sitting in the cloud, and they're, they're mostly dormant for most of the time. But if you have, like, actual smart technology, processing input from sensors 24/7, I think it changes, it certainly changes your resource needs. And it may be the thing that brings you further along in Compute need space to move you along in the cost curve, that starts to make things more make more sense in on prem. By the way, I'm gonna, I'm gonna push on this idea, it's gonna sound like I'm wildly Pro on prem. And I think it's more that like, just recently, my world has shifted a bit because of some of these big changes. I don't think I'm a natural sort of cloud computing nerd, right. Like, that's, that's, I've built a career on moving people to the cloud, like, don't get me wrong, it's, it's awesome. And I'm super, super geeked, about cloud technologies. It's just, I think this is shaking things a little bit.


06:03
Right. I think some of the terminology that you and I have used recently in these conversations is really the cloud is the incumbent at this point. It's kind of like the accepted common practice. I think what we are all trying to do is question assumptions, right? And question some of the ways that have become commonplace in terms of applying technology to make sure that we're not missing something, I think it's worth pushing on your point about cloud being less expensive, because there are some nuances to that, aren't there?


06:32
Yeah, 100%. I mean, so it's always like, if you think about two competing costs, moments, I think are like the points of, of the financial aspects of, of this kind of tech. In the case, cloud computing, you're paying a regular ongoing cost. And that's a little higher than you would for like, basic electricity and stuff like that, if you had it all. In house. It's notably higher, I suppose. And, there's almost no adoption cost, no change costs. So I can, I can sort of scale up, scale down, and I'm going to feel the effects of that I'm going to, I'm going to save or spend money based on the demand at the moment. And that's what a lot of us wanted, we wanted that sort of burst capacity, we want to, we want it all, in the in the on prem case, what you have that's like slightly different is you have a really high sort of initial capital expenditure, you're gonna have to go buy servers and get them in place, you're gonna have to, like create data centers, or what have you. And that's going to be slightly more expensive. The turnaround time for new hardware, you got a lag in terms of the access to that kind of stuff. And that's all different. But, the ongoing costs are usually in administrative manpower. So like, you're gonna, you can't run a data center, without somebody to run it. I mean, that's just like, maybe the, I suppose it's obvious, but, you know, with the cloud systems, you're, you're really pushing all that on to someone like AWS, or Google or, or, or Microsoft. And, and that's what you're paying those sort of like, hour by hour premiums for is to subsidize some admin to work the AWS data center or whatever. And so, I think, the trade offs as you go along this cost curve, that there is eventually a point in which that that large initial capital expenditure is, is small in comparison to the total cost of ownership. If you look out over like I'd say, a 10 year five year period, that becomes less of a big deal when you're thinking about, you know, cloud usage costs of, you know, millions and millions of dollars. This, it's not, it's not unusual for someone to spend, you know, 10 $15 million a year on their AWS bill. You can buy a lot of servers and hire a lot, a lot of people to run them for those kinds of dollar figures. And it just becomes non obvious, I think.


09:31
Me, let's talk about the other dynamics as well, right. I mean, there's a performance question that needs to be addressed. And I would say, coming back to our starting point is as relates to the machine to machine learning component, again, performance, what's the implications there? But also, from a security standpoint, what are you kind of weighing or trading off as you think about the value that this machine learning up occasion is going to have and how you protect that. And when it's also turning out and producing?


10:04
Yeah, I think many people would love it, if we could slap a one size fits all thing on machine learning and the cloud and on prem stuff like that. And the sad truth is, you can't, because machine learning isn't all the same. You know, I think, what you're, what you need to do internally is, you really have to figure out that computational profile of what this is. And so machine learning can take lots of different forms. But some, some, some tasks, you have large, really intense burst training exercises, like for the models, and that needs certain, like I compute capacity. And that, that can be a different, really different profile from the execution of the model and how to make decisions day to day. Or there are some things that are of the opposite profile, you have nearly no initial training and execution takes quite a lot of a lot of capacity. And so I think, again, filling it out that, that profile in terms of burst versus like steady state, I think, if you're an executive making these types of decisions, I think you want to understand those two different modes that your computational needs are going to hit. And so that, I think that's how you'll think about whether that initial cost works for you. And then if the scaling patterns are something that you can, you can manage, internally, because obviously, the more bursty things get, the more the cloud makes more sense. 


12:01
So let's bring this back and try to be as relevant as we can for our audience of people who work specifically in the industrial space to think about manufacturing. Where do you think some of the more bursty types of applications are likely to come up in an industrial process that will lead those folks in our audience to think about the cloud implications there?


12:22
Yeah, I think, for me, the ones I've seen are things that are more in that model training, that was one that already came up, I think the sort of like, usually, it's actually what we would naturally cause to be like, nightly activities. So things when the activity in a factory or something like that is actually low. There's a lot of, you know, off hours, computational needs, maybe that's things like reporting, or whatever, if they're batch reporting that needs to happen. Maybe it's daily analytics, or like, you know, that kind of thing. But those are the types of sort of batch bursty things that I see. Trying to think of, are there any other ones that are in the front of my mind? I mean, there's probably the meta, sure. 


13:22
On the flip side, the steady state applications, what can we connect from like a manufacturing industrial space into that more steady state type of type of need?


13:32
Yeah, I think I mean, I think the big one that I am talking to a lot of folks right now is sort of data transfer, like we're talking about a relatively escalated amount of data. And so literally just moving bits and bytes around to the, to the right places to even have computation to happen, or be in a place where reporting can get done correctly, or what have you. But it's that anything that's like, sort of like real time, which a lot of these these things are, I mean, we certainly often jump to that use case of real time, historically, is always the more challenging thing, because I have to have that steady state computational engine up, up and humming. But, you know, it's common that not everything needs to be real time. And so some of that thing, some of that stuff can get shoved to the, to a background, a task or a thing. That's, that's, that's done nightly. So I think, yeah, if you can imagine like, just for an example, if you can imagine sensor data coming in from your machines, and you're looking for anomalies and moments of that are expressing potential failure. So maybe that's, maybe that's because you have temperature sensors and as they get, you know, too high or too low. You want to get alerted and all that stuff. Those anomalous detections that, like a machine learning algorithm might alert you to are, are good examples of things that need to be pretty real time, right? Like those, those need to be sort of constantly processed and run. So they take up a lot of resources.


15:22
So if we were talking to like one of our partners in the industrial space, who's already all in on a cloud solution, are there any circumstances where we might consider advising that a step towards a hybrid approach might make a lot of sense, and might give them certain efficiencies or additional capacity in terms of what they're doing?


15:44
Yeah, I think, I think that should always be considered right, like, first off, everything in our world is very much so like, let's take some steps, right? Like, it's not like, let's, let's jump all in. I feel like the past innovation is being able to have a smooth curve to adoption. But I would say like, I say, first off, I'd say a lot of this can be done in a hybrid way. The big, there's, there's one big challenge or like to, I would say, two big challenges in our way, right? So one is, when you don't take the big plunge, you're, you're basically accepting all of the challenges of both sides. So you've got, you've got double the problems, and that that can be things like, we haven't talked about it, but like, if you think about, like, the reliability profile here, right, one of the one of the reasons to move to on prem, it would be your, when you think about uptime for your your tech or whatever, whatever it is, you're your servers, and all that compute infrastructure is gonna go down at very likely at the same time as the rest of your factory or whatever, if it's close. So if there's a giant power outage, you don't care that your reporting isn't working, because, dude, the machines aren't working. And so like, I think the tying the those two things together, make some sense to have some, some companies, some might say the exact opposite, which is like, when it's down, I need I need the mothership or something like that, to be able to know to be able to respond, and that's reasonable. But if everybody's kind of on site, that would be obvious. Maybe that makes sense. Whereas when you're in the cloud, you're worried about an AWS regional outage or availability zone going down or something like that, you're worried about some of those big things. And those have nothing to do with the rest of your business. And so like today's most industrial, you know, businesses aren't wildly disrupted by major cloud outages, although more and more that's becoming an issue. Cloud tends to be more reliable overall, but it's decoupled and its issues are decoupled from the issues of your business. So I think that's sort of worth saying, when you're thinking of a hybrid. The second thing, when you're thinking about perhaps pursuing a hybrid approach is the expensive data transfer, moving bits to the cloud and out of the cloud. If you're a heavy cloud user, you often don't think about ingress and egress costs that come from using a cloud. Because as long as it all stays in the ecosystem, they're happy to keep it all in there. And they won't charge you for a ton, a ton of movement there. But if it's going back and forth from your, your, your personal data center, or your factory or whatever, depending on the volume of that, and these applications tend to be higher volume than historically we would be thinking about, you've got to, you've got a sizable expense there. And that can be that's mentioned, just the sheer time that it takes like, it's hard to move. petabytes of data, just like that's just like physically can't happen in an instant. So


19:22
it's really, it's a really thoughtful way of thinking about I think we've covered a lot of the details around here. The one thing I'd love to ask you to kind of like is to put a bow on this, like as you think about this topic, you think about the space. You know, at TSI, we always use the word curiosity a lot, right? So as we, as we've been talking about this space, on prem versus cloud, the implications for machine learning implications for the industrial space, like what's the thing that you're really curious about on this topic, that you are kind of spurred to kind of like keep digging into and want to have some questions further answer about like, what are you curious about here?


20:02
I mean, maybe it's just because I got to speak with a passionate individual, but like, I'm really, I'm really interested in what changes that cost curve that those intersections of cost make sense for on the on prem versus the cloud. And I think books like oxide, people who are pushing the boundaries of, or where that sort of enterprise tech can sit there, they're just, you know, historically, like, Amazon and Google and all the like, are spending a lot of money and making those servers cheap to run, right? That's like, it matters a ton to them. And historically, that wasn't that type of innovation wasn't accessible to even the standard enterprise. So my curiosity is very much so in what kind of magic can folks like that make? Because honestly, physical, like hardware stuff, it for me is slightly more distant to the work that I've always done. And anytime we take large leaps in, in the physical hardware, I feel like it's like magic. It's like, it's like a whole new world opens up like, I still wouldn't even claim to fully understand quantum computing. But I'm just like, wow, that's science fiction. And I think we're actually on, like, people are certainly working on making science fiction or reality and in this space, right, right now.


21:36
That's awesome. All right. So we are going to use this theme of curiosity to lead us into what I am expecting is going to be a recurring theme or recurring question in this content series that we do. I would like each of us to share one thing from the past week. That was the best thing we asked of generative AI. Alright, I will go first. I'm getting ready for a big family road trip for our summer vacation. We have a quick, quick stopover on the way home we're doing just one night in Richmond, Virginia. And I was like, Jared AI, if I have just one day in Richmond, Virginia, what are the things that I cannot miss? So I was really impressed with how quickly he's able to give me a list of the top five gardens to visit areas of town to eat. I was like that was way faster than any Google search I could have done and it was pretty thoughtful.


22:35
And what's the number one? Give me that? Give me a good Come on. Give me the goods right now. I'm like invested in Richmond, Virginia all of a sudden.


22:43
The number one is basically an area in town called Hollywood known for its cobblestone streets and great vibe, apparently. So I'm going to pick up on that vibe. To a great extent we can while we're in Richmond overnight, and in two weeks time. Beautiful. Beautiful. What about you? What was the? What was the best thing that you asked of generative ai ai this week?


23:08
So yeah, like, mine is? Okay, so those of you who don't know me well enough, you're just gonna you're gonna get to know me through the podcast. But my personal passion is definitely in board games and a whole bunch of nerdy stuff. And when I sit and dream big dreams, I imagine quitting all this nonsense of being an engineer and opening a board game store. And I asked, I asked Chat GPT, hey, give me some names for my new board game store. And you're just sort of like, I don't know, pushing it. And like, they were great. So they're all we ended in a place that were like, 10 that were all super pun related. And I thought it was great. So we got things like The Board Room. We've got Decked Out Games. That was great. Board Stiff. The Dice is Right… come on. You'd have to pay a company for this one, like so good. So that's what I got. Oh, another one that I liked was Let's Get Bored. I'm not I'm not gonna make a place that's called Let's Get Bored.


24:24
I think The Dice is Right is definitely the winning answer there. Among those. I know that we probably have a bunch of listeners in the Richmond area and I need to do a quick correction before we go any further. Shockoe slip and Shockoe bottom are the two areas with the great vibe that I was meant to say. Whereas Hollywood Cemetery, which while may sound morbid, is apparently very notable for its historical significance. Not the same kind of vibe though as the other places so not something that you'd want to get confused about. So to our listeners and viewers in the Greater Richmond area. Apologies for me You can do that gray bear no pun. I mean,


25:02
That's the thing about generative AI. It's not always perfect. You know, sometimes you show up to a place call Hollywood and it happens to be a cemetery. No big deal, right? Like you just kind of roll with it. 


25:13
That's exactly what you have to just roll with it. That's the other name for your board game.


25:17
God, that's good. Like, I'm signed up. See? The chatty videos know about these things.


25:25
All right, I think we should end it on that note. Listen, I wanna thank all of you for joining us today. If you enjoyed our content, please subscribe. Our ambition is to continue to cover the important topics and trends shaping industrial innovation, and we will see you next time on The Modern Industrialist.