Rebasing Your Wobble Detector, Industrial IoT and Pivotal (Ep. 40)

Episode 27 · 54:55 · SRT · archive.org · all Tanzu Talk

we're recording on Halloween, and I guess we'll put it in the not-so-secret back end so that people can listen on Halloween. Now, that is, last I checked, October 31st. Now, here's a question I have for you all. You can choose to answer it however you want. What Halloween costumes are going to be in your domiciles today? And as an example, I don't think my wife and I are going to have a costume. We haven't really discussed that. Our son is in first grade, so he's had to have two costumes, so we're costumed out. But my daughter is going to be, if you're familiar with the Wild Kratts, she's going to be a Viva in an octopus creature power suit, and that's exciting. And then my son's school costume is Jacques Cousteau, which is fun. And then his regular costume is he's going to be a scorpion. So how about over in your neck of the woods, Richard? What's going on? Yeah, my daughter's about to turn three, so getting her to wear any costume has still been a struggle. So we've got both Wonder Woman and a princess, and she's been wearing them around the house. I think we're in good shape. The downside is it's going to be like 45 degrees tonight, so both seem like four choices. Yeah, you know, I think maybe at the three-year-old level, a good costume is clothes. Right, yeah, a shirt right now is a great outfit for her. And then my eight-year-old son's going to go as the Grim Reaper, which will also go well because it's supposed to rain tonight, and nothing looks scarier than holding an umbrella. And then this is an opportune time. Let's bring in our guest here. Why don't you introduce yourself before you answer the question? Sure. So, hey, everyone. This is Saurabh Gupta. I'm an advisory platform architect at Pivotal, working out of the New York City office. And I just help customers' current end perspective with all things Pivotal, especially on the technical side. So when your prospects and customers call you up and say, hey, I don't have a Halloween costume yet, what are you advising them on? So we start off by doing some pairing on it, I suppose. That's always a good way to break

problems down at Pivotal. But, you know, there's only one type of costume I personally know, and that's nerd. But that's a power suit these days. So you have to kind of... I was going to say, is this like 80s nerd with a pocket protector and taped up glasses? Or like, how has it evolved over the past, what would that be, 20 years or so? Oh, yeah. It's like a hipster nerd now, right? Like, these are all, you know, shorter, those ankle-high pants with the funny socks, you know, artism shoes. Yeah, yeah. It can get expensive to dress like that these days. You know, it's funny, the hipster nerd, as it were, I think the only thing it shares with the 80s nerd is the high pants and socks. I think both of them had that. But they were white socks in the 80s. They were absolutely monochromatic, bleached white socks in the 80s. Now it's like these absolutely rainbow colors. And these socks, no doubt, are delivered on a subscription basis weekly from some sort of service with a funny name. And they probably have silver in them. That's right. Socks as a service. Well, I think it'll be an exciting Halloween here. My kids woke up just bouncing off the walls, as it were. And so I was thinking, like, I know what we should do this evening. Load them up with candy. That'll go great. How much of that candy do you end up eating? You know, I actually, I don't want to be self-aggrandizing, but I actually don't like candy that much. So I don't actually eat it. But I'll probably, I'll tell you what, if there's one of those, those little butterfingers in there, you know, have you seen the man, I, for someone who doesn't really like candy, I'm a fool for butterfingers. I love those. I feel like I can feel them like drilling into my teeth as I eat them. And it's good stuff. Also, for some reason, some, you know, I book all of my own travel. So whoever, whatever stupid person books my travel booked me on an early morning flight tomorrow. And I don't know how that's

related, but, but we'll see what happens. Be very excited. Yeah. You should get up early with that. Yes. But I'm going to be narrowly missing by a day. I'm sure they're doing some hilarious, spooky, like safety talks on flights today. And it's just, just such a shame. I'm going to miss all of that. I don't know. Things, things have gotten pretty sedate lately. Maybe, maybe there'll be a Samsung galaxy seven demonstration. Exactly. That's the scariest part of the plane demo right now. Yeah. If you dress up as one, will you be allowed on the flight though? That's right. You can come melt our chocolate for the free hot chocolate we're giving people in first class. Right. Making s'mores. Well, speaking of industry news, before, before we get to the second half of our show, I wanted to bring our guests on to talk a little bit about, you know, stuff we do, we pivotal do in the industrial world and the internet of things and, and stuff like that. But we've got a little bit of news items to go over at first. I think, I think there's a, as always, just about every time we talk, there's some interesting partnership acquisition thing going on and a few cloud related things. So why don't you kick it off for us, Richard? What's, what's, what's catching your eye over there? Well, there's earning season. So we've seen all the hand-wringing and celebratory information about, you know, Apple did fine. They think they had earnings drop in general, which was unusual, but still printing money. Microsoft's doing well. Google's doing well. Amazon, again, had a good quarter. I think it was less than expected. So they saw a bit of a stock drop, but it's also seems to be executives shifting between jobs season. So Adrian Cockroft joining AWS, him, Caroline McCrory joining Microsoft. Just it's that season. I think people get through some bonuses and they get through the summer and they rotate between positions. So always interesting to see that moving and shaking. And then my alma mater CenturyLink acquired level three communications this morning for lots of billions. So always a end of year sales apparently

going on in the, in the enterprise world where companies are buying each other just before Black Friday. That I'm going to try to use that as a anti McKinsey title sometime on a slide, just in quotes for lots of billions. That's, that's a, at some point it gets to be silly money. So it's not even worth talking to numbers. So, so since it's your alma mater and I always feel a little awkward talking about acquisitions at places I used to work, so I don't, don't go all awkward or whatever, but, but like, what is, uh, what's up with CenturyLink and level three? Like I don't, I know CenturyLink, but I mean, how would you characterize both of their businesses? Like what, what do they do? I mean, I think this point it's, it's network, network, network. I mean, level, level three is, you know, has a lot of fiber. I think I saw, you know, CenturyLink's adding a couple hundred thousand miles of fiber now. And so, you know, you get a lot more connectivity to office buildings, to residential as, as companies like CenturyLink are trying to compete more with the Comcasts and Verizon's of the world for, for end residential users. But I think this is, you know, a move back towards pure telecommunications for both of these companies. It's a, a giant company now that may be able to take on some of the traditional telcos, but I think this represents a bigger shift back to what they were new and were comfortable with versus some of the more recent flirtations with more cloudy stuff. Hmm. The pipes, the pipes. Meanwhile, AT&T wants to buy a Time Warner going in the opposite direction, I guess. Well, and Google stopped doing their Google fiber stuff. Oh yeah. I saw that. Fiber is expensive. And so these things are tough and it's a capital, intensive business. There's only a few things that seem to be really hard to enter nowadays. And it's either making planes or laying fiber. Other than that, everybody can be disrupted. And, and of course, this is going to skate on the edge of my knowledge of the world, but you guys, don't you guys have a CenturyLink stadium up there, like up and up in your head? Oh yeah. My, I'm sitting in the office, uh, right across the street from CenturyLink field, which is home of the Seattle Seahawks who will finally be Superbowl winners again this season. Now I'm hoping that it has more than three levels and in the future, whatever level three is,

has some sort of special denotation on it. Just something, you know, for the employees to appreciate. You would hope so. You would hope so. So also, uh, last week, you know, I didn't actually follow up on this news, but there's, there's new MacBooks out, right? And didn't they put like a touch bar on it? Was, I saw a bunch of jokes about them removing the escape key. Did that actually happen? Or, or I think so. I, it's just funny because I mean, Microsoft had an event that day before. And if you, again, for following my Twitter feed or other people, it was the Microsoft event was, wow, this is really some neat stuff. Maybe it's not all practical. Am I going to spend, you know, three or six grand on a giant monitor that turns into a screen that I can write on, whatever. But it's a very creative play for surface studio versus the Mac event was, it seemed like pretty widely panned as, gosh, we waited three or four years for, you know, a touch bar and some more memory. Like what else was going to change here? Not a touch screen, not these other things that people might look for from a MacBook pro. So, Hey, good news. Things, things did evolve, but they're killing more of the ports on the side and, you know, adding a touch bar, but not a touch screen. So like, okay, not, not bad, but you wonder if Microsoft's at innovating them at this point. Good old surface. Remember the surface table? That was, that was ahead of its time. I like that. Yeah. That was something you see in those old arcades or the old pizza parlors where you get to play across from each other. Yeah. Yeah. You know, it did have a little miniature rant. Now I got one of those, uh, iPhone seven pluses, which is a great phone, uh, as my, you know, it's, it's good stuff, but it is the, uh, the prediction about the, uh, the lack of a headphone jack is a little tedious. There have been many times over the past since I've had it, uh, so to speak, where I, I, I think, you know, I was traveling last week and I think it was like a day trip. And I think three times I sat by a plug and I was thinking like, cause I want to charge this. And I was like, Oh, right. I'm actually listening to something cause I had it wired in. So it is like, uh, I would categorize that under not cool, right? Like it's, it's bearable and sure I should use Bluetooth, but like Bluetooth, like a disaster, like it's

just really annoying. Uh, so it is, uh, you know, that's, that's an annoyance. Like whenever a ports are removed, uh, it's, it's kind of not good, but. And I understand that to try to move to progress, but when you're yanking out, you know, HDMI ports, like that's how I plug in my monitor at home to my Mac book. So what am I going to do? Buy another dongle? Like the dongle market is going to explode, which is maybe the title of this, this podcast now, but you know, it's just, do I want this many appendages to my computer or should my computer just do the things I want it to do? Yeah. Start investing in big dongle. That's big dongle is going to take over the industry. You watch. Yeah. Or at the very least have a dongle Halloween costume. Oh, see now that's why you're an advising platform person. It's, it's power without accountability. My friend, it's a great place to be. Yeah. Dongle costumes. That would be hilarious. Well, I, uh, so, so what is the surface studio like for, uh, for like, uh, graphics pros or like what's, what's the, uh, is it similar? What's the, the iPad pro? Is it a similar idea that this is like a work item that you would use? That's what it looks like. I mean, it's again, it's a big, almost all in one computer that can also turn into a writable screen and this and that. So Microsoft going hard against the creative audience, which is, you know, micro, uh, Apple's home turf and arguably doing it better right now, just given their focus. And if you look at, you know, Apple's earnings, they don't make much money at this point on the computer market. They make a heck of a lot more on music and software and things like that. Right. And iPhones. So I guess you can forgive them for not paying as much attention to it, but you just wonder if they'll, they'll actually start to lose some of the vibe of being the best platform for creative people. Yeah. Yeah. Yeah. We'll see. Yeah. I noticed that final cut pro is now only $300 in the app store, which is, uh, back when I did video editing, it used to be like, I don't know, $2,500 or something. It's crazy. Uh, well, I had a couple of, uh, little items that aren't really like newsworthy, but I wanted to point them out. Cause I think they're, uh, they're highly relevant to things we talk about and things we talk about. So the first one is a

bit of a, uh, would you call it a jeer? I was reading this piece about, uh, our friends at IBM and how like Watson is great for them and happening. And there's this little mention that IDC had predicted that, uh, by 2020, uh, the, uh, let's, I can't do the math, but the, the, uh, the AR market was going to be something like $45 billion or $46 billion, which on the surface of it, like, you know, you know, if long time listeners know, I'm always like, what's up with AI? Like, could you just help me schedule meetings across organizations? That would, that would be good to have some artificial intelligence in that. But what's sort of astounding about this, and maybe I have it completely wrong or I've done my math incorrectly, but it's, this is 2016 last I checked. So 2020 is less than three years away. And then if you read this article, it's basically like currently the, uh, the AR market is like seven or $8 billion. So somehow over three years, there's going to be almost $40 billion worth like generated on AI stuff. So, so listen, there's going to be some gigantic steps that obviously we haven't, uh, taken a, taken a gander on that are going to be happening. So we should just get ready. It's going to be awesome. There you go. No, those are, those are big numbers. I mean, you do see more software providers, obviously baking AI in, but it's amazing to see it getting that big. I mean, that amount of progress over three years would be, would be fantastic. Right. But that's a, that's a big job. And then, and then there's another, another, uh, more nerdy, uh, sort of mixed thing. I, I think this is an older study if I remember from 2014 and it had been updated recently, but I, I came across this, uh, academic study. I forget who it's from, but academics. And they were studying the effect of Airbnb on hotel pricing in Austin and Dallas over the course of, I forget how many years. And, you know, the, the, the headline grabbing top of the thing was that basically to nuance it a little bit, basically if you were in the lower end hotel market, what we used to call motels as, as I recall, but the non-business traveler, uh, luxury market, you can expect your

prices to drop like eight, eight to 10% because Airbnb is in your market for whatever science reason. Like I love reading academic papers because like most of them are methodological stuff with funny Greek letters and I can pretend like I'm smart, but anyways, like it's a, so there's established that, but I think, I think what's really interesting for people who listen to this is, um, uh, there's a, there's an explanation at some point in the paper going over the, let's call it elasticity and lack of elasticity and in, uh, the inventory of hotels that beds, I think they say. And basically it says in a traditional hotel business, it takes about four years from idea to opening the hotel doors to build it out. So if you want to increase your capacity, uh, it's a long time versus what's interesting about Airbnb in this discussion is that they can basically do it, uh, instantaneously. People who are renting out their houses or rooms can either list them or delist them like on their whim. Now, of course they can't build a house on, on their whim, but the idea is that you've got this infinite supply of beds that you could be in. And because you've got this, all the software and the system and the business model enabled by all that software between Airbnb, you like get around this constraint, this four year constraint that existed in the hotel market. And it's just, it's an interesting way of thinking about a lot of the, uh, kind of small batch thinking and even kind of cloud dynamics that we're always saying businesses need to, uh, get onto that, uh, is explained pretty well. And, you know, eight to 10% ain't, ain't, uh, ain't nothing to shake your head at when it comes to reducing your top line revenue. So it, uh, it does point towards to wrap up my rambling here. Like in my mind, I filed as way is one way to look at, uh, all this software defined business cloud native, or let's finally do software and in a good way stuff is what are constraints that your current business model has that just seem impossible to break? Like it takes us four years to build a new hotel. So we had to plan four years and you should

go through those constraints and ask like, is there a way that we could use software to change our thinking and enable a new process that would remove that constraint from us? Which is, you know, it's a good way to, uh, set the course of your strategy. Right. No, I like that. That's a, that's a great example. It's a more out of the box and just always looking software, software tech industry, but that's a, that's a cool example. Yeah. So there you go. So we'll put links to those in, uh, in the show notes there. And so now, so now let me reintroduce my guest. Hello again. And, uh, or our guest, I should say. So I, I wanted to have you on because, uh, over the past couple of weeks. So, so someone had the bright idea. Hey, let's go get Coté to talk about something he doesn't know that much about, uh, which is always challenging. So, uh, you, you were like, you were like my secret professor on all of this and, and you helped like, uh, train me up on, on what, what, uh, some interesting things that we do in the industrial space and internet of things and so forth and so on. But as we'll get into, like, I think for me, the thing that was most interesting is that, uh, you had come up with a, the way that I describe it as a way of applying small batch thinking to IOT cases, which is to say, what is like a process by which we can introduce a feedback loop, uh, into industrial stuff that we actually do something with it. Right. And, and, you know, as the jokes always go, whenever you hear about internet of things, it's always sort of like, uh, I think Harrell from GE puts it this way. It's, he's always like, ah, there's, you know, blinking toothbrushes and thermostats, you know, like there's all this consumer world stuff, but, uh, also to crib another phrase, uh, phrase from him, right. There's also like the big dirty noise machines, noisy machines. And that's where, um, you know, a lot of what pivotal does operates in that area and where there's a lot of interesting, almost kind of hidden stuff going on that it's all, it's hidden by the fact of it being, I guess, boring to most people, right? Like, like managing turbines and agriculture and all this

stuff. They have really interesting effects, but the, uh, what's happening there is, is not so exciting as USB toothbrushes, I guess, but yeah. And it's all these names, you know, with, with fancy, fancy names with vowels missing and, you know, fancy new top level domain names as, as endings like .io in many cases, right. That, that are trying to be IOT companies, but the surprise to most developers even is, is that there's really a lot of truth in the GE ad where they're saying we're an industrial company that is also a software company because industrial companies have in, you know, even in the analog world really done this for a long time. Um, when you look at how the electrical grid works and some of these, how these major systems work, how, how power plants work, there's a lot of, um, custom stuff that had to be done to, to continuously monitor things at very, very large scale. And, you know, I always get a chuckle out of folks doing cute things by combining different Arduinos together. Oh, I did an IOT project. Yeah. It's a connected thing project, which is not an IOT project. And, you know, that was, that was something that you and I, we were chatting about last week as well. Yeah. And, and, and so, so keep, uh, I was about to lead the witness as a word to get you to say all that, but, but you led yourself, uh, the horse and the water and the whatnot, but, uh, lay, lay some more of that groundwork. Like what, what are some examples of, of let's call it industrial IOT or whatever? Like what are, what are like the machines that were previously like very analog driven, but now they either you can rapidly get data off of them or, and, or they have computers embedded on them or they're like hooked up to the network as it were. Like what are, what are some interesting, um, examples of that? Right. So, uh, so let's actually kind of take one step back and, and go over a lot of this, uh,

glossary in a way, because there are many different ways in which people interpret IOT. So when you go to, you know, the tech crunch website, or you go to some of these tech blogs, you have, oh, here's an IOT company that is, you know, creating these cute little things to teach children about coding and all that. Like, you know, not to, not to poopoo any of the work that those companies are doing. Some of it is very, very fantastic work and very essential. The, the fundamental difference that, you know, or even when you take in hue lights or some of these, you know, a wifi enabled dimmers and stuff, what you're really replacing is a wire with a wireless protocol. Basically that's what you're doing. Oh, instead of there being a light switch that I flick on and off with human intervention, I can push a button on my smartphone, which will essentially convey a message to something which will then electronically cause that switch to flip, which will turn my light on or off or, you know, increase my, or decrease the intensity. That is just a connected device. It is not necessarily IOT because what you need in IOT is a decision-making loop where there is a cause and effect. And if you look at how all of industrial machinery operates, it is designed to be in a feedback loop. There's the homeostasis that is continuously monitored where, and, and you have these analog feedback systems that are essentially sampling the system continuously. It's, you know, the, the frequency, the sample frequency is, you know, either hundreds or thousands of times per second, because when you have this massive machine that's burning fuel and it's spinning and it starts to wobble, you have to really very carefully figure out whether there's an issue here or not. And very, very quickly. So for there to be, you know, you know, when you have, when you have an electronic doorbell where you press a button on your smartphone and a message goes up and it, you know, bounces off a data center that's 2000 miles away and comes back through your wifi router. And so you're essentially making a 5,000

mile round trip to cross three feet. That's not IOT. Like that, that would actually be an anti-pattern. Yeah. It's, it's, it's like the plastic bag of cyberspace where you have to travel all that way and be, yeah, you know, it's, and, and, and this, this, this will kind of reveal, uh, my complete, uh, lack of industrial, uh, hardware knowledge being a pure sort of like software person. But it's interesting in studying a lot of IOT cases, it seems like one of the Uber things is you, you alluded to alluded to it is like wobbles bad. We've got to detect wobble. Like whether you're doing drilling or turbines, but I guess a lot of most many things spin very rapidly. And there's, there's almost like if there were, if there becomes a, uh, uh, design patterns of IOT, I'm sure one of them will be like wobble detection. And, uh, yeah. Yeah. I mean, any, yeah, if it wobbles, it, it's not going to last long, no matter, no matter what it is. Um, even, even, even if it's not spinning, you look at a human, a wobbling human. Yeah. You got to have it looked at. So, so basically in, in all of these, and, and a lot of these machines have been connected online for a long time, you know, they were physical cables before the internet. So what the internet does is break up a signal into packets, but, and then figures out a way to route them. But if you don't have that system and, you know, this is, this is Edison's era, this is, you know, the 1940s, you're building these massive machinery plants and, and, and, you know, refining stuff and building stuff and turning, uh, rubber, coal and steel into, into cars, really, that's what's happening in Detroit. Uh, how are you monitoring this stuff? So, so there, there is always this loop where we will extract data. The data is going to be some kind of a, some kind of an electrical pattern, which we will then put through some semi electronic device. Um, you know, whether those are resistors or capacitors or some other sort of like simpler, non microchip style things, and then create some indication of

how we control the input into that and, and how those two things start to get more and more separated. So then you, you have a control panel, which is again, starts to get more and more removed so that you can bring in more scale. So you have one central sort of mission control style system, and then you have plants in different places. And then the distance between those plants starts to increase. And with increasing automation, increasing sort of computational sophistication, you can get into situations where in Canada, you have completely almost autonomous oil, uh, refineries running in the tar sands. And you have, you know, the, the engineers that are monitoring it sitting comfortably in Alberta in a, in a heated building and going home to their families at five o'clock, uh, that sort of thing. So, so it enables that and, and companies have been doing this for a long time. And that's really a surprise to a lot of folks who are trying to get into IoT is that there's so much that's already been going on for so many years, um, that these use case specific things, um, you know, the, the, what we deem the classic quote unquote dinosaur industry really has a leg up just in terms of size and scale to tackle some of these problems. Can I ask you a question about that? So, I mean, are, are the industrial companies are using this, are they retrofitting to collect some of those, that information or how, how are systems, you know, modern applications or rather systems collecting some of that data coming off of these things? Are they modern pieces of hardware or are they, how does that work? Yes, that's, uh, that's, that's actually coming from the OEMs. So, so that's a whole, like, there's, there's this, there's a whole industry that does nothing but these process control loops. And, and Pivotal works very closely with some of these major industrial manufacturers that do one of a bunch of things, either they make the machines themselves or they make these process control systems. I see. And, you know, there, there's an increasing sort of digitization of this, but fundamentally there's, there's, um, there's sort of three, three steps to this. You have the thing that is

producing this output. You have something that's taking that output in. You have some, some other mechanism where a decision is being made about what to do with that sort of received output. And then something that is transmitting an instruction to change state, which could even be an instruction to maintain state, but some, some instruction to, to the machine to, to maintain, manage, monitor, whatever with their state. Now, these are, these are classic industrial control systems and they've been analog for a while and they're getting more and more digital as we go. The question is, is correctness and security. And, you know, where, where do you want to put in, um, you know, the monitoring and where do you want to do the data capture to record it somewhere? And how do you actually then store that analog signal for later use? So those are the, like the answer is yes and no, because it's very, very, very use case specific. If you go into highly regulated environments, those things don't change very well or very often because you need to put in a lot of certification before any change can occur in more newer fields as, you know, as more, um, uh, you know, if, if it's a newer technology, then it will be built digital from, from the get go, but then it'll have to be sort of as that. So, you know, I mean, the, the regulatory burden here is significant. So, you know, to, to sort of, uh, uh, teeter totter between the abstract and this was abstract and the specific, um, uh, what, you know, I, I was kind of going over that Airbnb study earlier, or I don't know if kind of, I was shallowly going over it. And, and one of the things I liked about it is it sort of like exposed an interesting constraint that, that you can, uh, exploit or limit yourself by. And in, in so much that many, many industrial things have been connected and, and at least had some like, you know, uh, controls in them, like, what are, what are some in, in more of

an IOT era? Like, what are some of the constraints that get removed, uh, that allow you to do something new and interesting? So if, if you're, if your tar sand buddies have been up there, like remotely drilling forever, like what's different about stuff now that it's something to be all agog about? Right. Yeah. That's, that's actually a great question. So, so let's actually make it, make it more, um, uh, more relatable. So case in point is you have these, uh, utility meters or, uh, you know, that, that, that folks come around and, you know, read from the outside, the, the utility workers come and read, read their meters or, you know, do some work with that. Um, and then you say, oh, let's make them connected so that we don't have to send people out to each house to read it. We'll just keep getting their readings, um, ourselves, uh, from central office. So there's, they put, they either put a cellular chip in it or they wire that to your telephone line or something, um, whereby the meter reading can be, uh, sort of, uh, extracted from a central mission controlled style, uh, room or facility. But what you've done really is just done a status reporting. So where this starts to get interesting and, and these sorts of things have existed for a long time. Another case in point is your, your electronic, uh, key cards when you go into a hotel. Well, that is that IOT. Well, it's an electronic doorknob, but it's not really IOT. Um, and, and there are many other situations where you have these connected devices, but they're not necessarily IOT. The, the, the significant thing about IOT is that you're able to capture the, the time series data from that device and are able to analyze that and use that analytic and machine learning input to inform how you control that device, whether that control isn't, uh, affected through a human means or to an automated means. But this control loop extending that stretching out this control loop by saying,

we are also going to inject decision database decision-making into how this thing is controlled, how its behavior is controlled. That's what is so exciting, which absolutely exist before. And, and, and, and correct me where I'm wrong or fill in things I'm missing, but, but the, let's say over the last five years, the, the, the major, the technological advances that make this possible is probably, and, and cheaper is spread across all of these. It always has to be cheaper, but, and maybe that's the only thing is, is you basically, you have cheaper networking, whether it's cell networking or whatever, or so it's, it's always possible to like get a dedicated line to your convenience store as they do. And you got all those satellite dishes up there, but now it's cheaper for everything to, to be online. And then also, uh, I guess just raw compute and storage is cheap enough that you could actually do something with that, that you could store it and do something with those devices. And then in addition to that, because compute is cheaper and I guess storage as well, you can actually do, uh, more analytics than, and you do them faster than you could previously. And it seems like, at least when I look at it, the, the upshot of like, we've always had connected doorknobs, so to speak to whatever beyond connected doorknobs is, is that you, whoever you is a controller, I keep thinking of Homer Simpson sitting in the atomic control unit thing there, like at the control panels. But so when Homer's sitting there, he can actually very rapidly, uh, do some sort of decision. Right. Whereas I think, I think the loop, the speed at which you could do analysis before was both expensive and a lot slower, uh, than it used to be previously. Right. And it's not just that, that Homer Simpson can do a decision, but he can actually get three options that there's a 90% probability that this is a meltdown or there is a, you know, 95% possibility that there's a pump error here, or the, these are your sort of choices because we can infer from this data that we're getting, here's how it matches with previous patterns. Right. That's interesting.

Right. So, so connecting your electrical meter to, to the utility is, is like low jack. It's like, where's my meter? What is it's reading? Where's my car? What's this odometer? Right. Very simple. Okay. Doable, nice, has value, but it's not really interesting. The fun stuff happens when your utility says, here are what all the meters are doing. So how the house next door is not using as much as you are. So how do we now create a demand response system whereby not everybody turns their washers on or dryers on at the same time? So we don't have to fire up a new turbine, which means now they're sending data back to the meter. Right. Not only is it, is the data being sent back to the meter, that meter now can communicate with the washing machine and say, don't turn your cell phone right now. Wait till the load next door is done or wait till we give you the green light because, hey, as an incentive, we'll actually make it cheaper for you to do that. Right. That loop, that closed decision loop that says we've analyzed terabytes, petabytes of data. We know when things are happening. We know what these patterns correlate to. There is an insight here that is actionable, that has business value. That's IoT. So when I see connect your Arduino sensor and have it send a tweet, very cute, very nice, but sorry, that is not IoT. Yeah. I'll put a link to it in the notes. I didn't appreciate this talk until recently, but way back at the 2015 CF summit, Harrell from GE gave this great talk about how they're using cloud foundry and various GE things. And the case he has of wind turbines has all sorts of good examples in it. But I mean, basically they, you know, industrialize up their wind turbine farms and there's all sorts of interesting, uh, things. And, and once again, wobble comes into play. Uh, but, but one of the more interesting things is, is that it used to be, they would have to schedule, uh, maintenance. They would have scheduled maintenance. Like, I don't know what the period was every two weeks or something.

They would just send people out to check all the turbines just because they didn't know if something needed to be serviced. So they would have to check it. But now, uh, you know, I like this example because it bleeds into the, uh, the meatware space, so to speak now, because they have like extremely rich data about wobble and other things, they can just send out, uh, basically maintenance people on demand, which has like a tremendous cost saving to it. You're not just like wasting rolling trucks out and everything when everything's, uh, correct. So, you know, I, I thought that was a good example of like, uh, a pretty simple, straightforward thing, but maybe not obvious that you can't do that unless you have this, uh, a bunch of connected things sending it to you. And then it plays right into the loop you have of like, now you can make a decision, which is to send out or not send out a truck to check on something. Right. And, and that's why it's important to have lots and lots of these, which is where scale becomes important. And scale is enabled, like you said, just a few minutes ago with, with lowered costs. So the moment you have a commoditized sort of, uh, substrate of, of very cheap storage network and compute, that's when you can say, okay, now this becomes interesting. And, you know, like, like it, it sort of makes me think of, you know, what if, what if you had infinite energy, what would you do with that? Well, that's when you, things get interesting because you can have things like the transport and replicators you see in Star Trek. But if you don't have that, then you can't do any of those fun things. And that's the same thing here, unless you have this elastic on demand, infinite or near infinite way to, to manage this ingest, this, this amount of data, this amount of repeated analytics from multiple devices. So you can create a statistical models because, you know, you can't ever make a model from one data point. Um, because if, if you could, then you would have one person in the country vote and the election would be over. Cause you could essentially use that as a proxy for many, many, many other things. Thankfully it's not that. So you really do need a large statistical sample before you can derive anything meaningful. But then with that large data comes the, the issue

of, of handling it at scale. And, and so, so along those lines, like what, uh, like, like, you know, so we work, we work to this other slide you have with the loop that's in my head. And there, there's, there's a division of like, I think you have different words for it, but like the hot data that you have and the cold data and, and then like, uh, some analytics and then finally building applications over it. But, but what are like the patterns of the types of, I don't know what to call them technologies. You have it at least at all the points on this loop. Sure. So actually let me, let me, you know, for the benefit of the listeners, let me kind of verbally sketch one of these out and try to imagine that you have a device sitting out somewhere and it's, it's producing this, this data, it's producing something, it's producing a signal and you're somehow capturing the signal. So that's the ingest. Now that signal is ingested and it has to be passed on to something that is reacting to that signal. It could be, um, a pre-baked algorithm. It could be a pre-baked sort of, uh, cutoff could be fused. It could be something very simple or something very complicated. It doesn't matter, but something has to take the signal and react to it. And that reaction then has to go back out to the, you know, the thing, whether it's the same thing that produces signal or something else that can impact that or, you know, something entirely different doesn't matter. But fundamentally there's the data production, data ingest, data reaction, and then data outbound, like the impact of it. And that's sort of your, and this loop can be very close to where the data has been produced, or it can be very remote, but it's, it's supposed to be very fast. And that's sort of your basic process control. Now, one thing you can do is once you've ingested your data, you can also put this in, in what's known in fancy terminology as a device shadow, where you're saying, how do I accurately represent the current state of my system? And, and, and what that's, what that's doing is saying here's an object model or something that's in memory. So it can be updated very, very

fast. It's not a high latency operation. And then it can be queried very, very fast. And additionally, then you can do some fun things with it, like a real-time analytics. So as I'm, as I'm putting data into it, I'm conducting a real-time test on that data sample to see if there's any kind of event I should generate or some kind of trigger I should, I should initiate based on the input that I just got. But you can theoretically update this thousands of times a second. And then you're passing this on. So you've done your, you've done your sort of hot data slash device shadow. You've done some real-time analytics and you're then you then have to surface that. So someone has to be able to see it on demand. And that's kind of your dashboard. So this is sort of the, the near-term loop. So you have the, the real-time, you have the near real-time, and then you have the, the last layer of it, which is the archival. So as you're, as you're sending data to, to the hot data, to the device shadow system, you're essentially splitting the pipe and sending it to your archival store. That's your time series database. So that's not just the here and now. It is also the past, you know, three, six, eight, you know, 12 months of data relating to your entire system. So this is your big, massive, structured, semi-structured, unstructured data store. Is that the historian sort of thing? Yeah, the data story. Yeah. And you can then use, you know, ever more sophisticated or even time-consuming technologies to scan this data every X amount of time, whether every 10 minutes, 15 minutes, whatever. And this is where highly parallel data analytics technologies become relevant. And again, these are the challenges because traditional data systems, traditional RDBMS systems, et cetera, they're not up to this task. You need fundamentally new types of systems that are built to be distributed, parallel, concurrent, resilient from the ground up. And you run those machine analytics, your big machine learning, your big analytics on that data. And then you can essentially do a couple of very interesting things. You can obviously surface those insights,

again, so your KPI dashboards can take those batch analytics insights. But you can also update any real-time analytics that are occurring. You can actually say, you know what, we've noticed that there's this X amount of drift. So you should really rebase your wobble detector by this little bit amount because we've detected that everything is kind of wearing down a little bit. So the tolerances are changing. So why don't you update yourself with this new data? And none of this involves any downtime. It's all continuous. It's all automated. But what you're now doing is saying we can take insights from batch analytics, from big machine learning, and inject them into a real-time analytics pipeline essentially instantaneously. That's a game changer. And then all of this gets surfaced. And then you can also, depending on the use case and the security requirements, you can actually integrate your outbound based on the insights and the triggers that just got created based on these machine learning models. So you can say anytime, you know, the machine learning says, yeah, this is okay, or we approved this model, why don't you just apply it and send these state change messages out to the edge or the device? And that kind of completes your whole automated decision-making loop. So basically, you can imagine it either as tiers or layers or concentric circles, whichever mental model makes the most sense. But there's, you know, starting from the edge or the device, there's a real-time, near real-time, and time series loops. And each of them has a different value and a different play. But from an IoT standpoint, what you have to do is choose which ones are the most relevant use cases for you. And then how you sort of solve for those, and then how you complete that decision-making loop and where that occurs. And then you come into the more pragmatic considerations of what's going to be my data fabric? What's going to be my ingest rate? How big should I make it? What's it going to

cost? Because even if, you know, even very, very small amounts multiplied by infinity become very, very large amounts. It's like Arthur Dent leaving the light on in his bathroom in Hitchhiker's Guide to the Galaxy, right? Like, you know, it's pennies, but over the age of the universe, it's a big amount. So costs become very significant. And there are some numbers that I ran where even if you have, you know, very small amount of devices out there, a few million, and they're producing data every X minutes, the amount of data you start to get is significant enough for it to be a Google-scale problem for your systems. So, you know, that's sort of the high level of how to think about IoT, how pivotal things about IoT. And these are non-trivial problems. And, you know, we solve them with customers. And customers are still our customers, so we must be doing something right there. The ultimate feedback loop, remaining customer. Yes. I mean, you talked about some of the tech. So yeah, I mean, tell me more a little bit about how some of the pivotal tech fits into something like that. I mean, you've touched on a bit of it, but, you know, when it comes to that whole loop of taking in data and making decisions and pushing things back out to the edge and all those components, where do you see that intersection? Sure. So let's start at the ingest. And you need something that's very lightweight, that's very resilient, that's built for scale. And, you know, one thing that Pivotal has in its technology toolkit is RabbitMQ. And RabbitMQ is a very, very capable system. There are almost an infinite number of permutations you can have it to and tune it to for it to meet the use case that you want it to. It's very flexible. It's the system that runs WhatsApp, for instance. And I mean, I don't know if it's the Pivotal version or it's the open source version, but there's some version of RabbitMQ that's running underneath WhatsApp. But it's billions of messages and very high reliability. And I recently

had a customer ask me, can you give us some numbers, some performance metrics for RabbitMQ? And the only answer to that is, what do you need? Because you start with what you need, and then you can build these things out to be the right scale to handle that traffic. But from a technology standpoint, it's a really good technology. So that's the ingest part. From a hot data part, we have, you know, the scale out in-memory data grids. One example is Gemfire, that is a Pivotal product. And again, Gemfire is this amazing product. It's battle tested. It's run in some very, very mission critical environments. And it has this aspect that you can actually add more nodes and rebalance your clusters. And, you know, it actually gives you scale out memory. And it also has the advantage that you can run process analytics on the inbound. As data is coming in, you can actually do an eval on it before you stick it into an in-memory table. So again, very mature technology, very high speed, very well understood and known and deployed, and operating very successfully in different places. Then you have sort of the archival data store, where you could use any number of technologies, you could use any kind of Hadoop distributions, etc. And if you're using one of the open data platform compatible Hadoop distributions, you can set Hawk on top of that. Or you could use something like Greenplum database, where you get SQL like access to your entire data store, but you are able to query it in parallel. So you're essentially moving little bits of query code and not petabytes of data to run your queries, which then makes batch analytics, large-scale machine learning very, very feasible. And then, of course, there are things like Madlib that our data scientists have produced. So these are open source machine learning libraries. And then there's Cloud Foundry, where all this insight becomes actionable as software. Fundamentally,

you have to take this stuff and do something with it. You can make a PowerPoint or a presentation out of it and give it to a meeting. And then that insight just goes away, because it's okay, that was interesting. Now what do we do about it? And the actionability comes from software. So software lives on Cloud Foundry. It is essentially the application dial tone, where even parts of the ingest can live on it. You can have different types of stream processing. But the fact that it transforms your IaaS into this elastically scalable application fabric with very little work on your part. That's pretty incredible. And then there are other sorts of applications you can write that are actually doing the outbound, or you can integrate with existing systems, etc. But really, you have to carve out the pieces that work best for the use case that you have at hand and select from a fairly good, rich, diverse, and mature toolkit to build it out and get speed to market. Because if you start to develop this from first principles from the ground up, you wouldn't go in and build your own Hadoop. You wouldn't go in and build your own networking IP stack. These are completely not value added things. But what you want to do is build those models, those algorithms where you actually have a demand response utility business running, where you can charge customers, where it's accruing to your bottom line. So this stuff helps you get started very, very quickly. Awesome. How do people learn more about this? Do you find that people are publishing some reference architectures and seeing how you piece this sort of thing together? Are people learning collaboratively, or is it just kind of organic as these systems start to collect the data? You start to emit the data. All right, now I want to be smart with the data. I mean, how do people take that next step? Even if they don't have these industrial cases, I'm assuming many of these things aren't just connected doorbells. You might actually have insurance and other reasons you're getting car data or what have you. So how do people get past the hype, in your opinion, and get into actually doing something that's meaningful and

helping them be more efficient? I think the most important thing is to start with a use case. Because what happens is, the business says, we want an IoT platform. And tech here is, oh, they want an IoT platform. And tech is great at building things, like in-house corporate tech. And they start to build out, quote unquote, an IoT platform. But that means so many things to so many different people. So actually going back to the business and saying, hey, what do you mean when you say this? What do you really want to do with this data? How often will this data be collected? What kind of insights are you looking for? Is device state really relevant to you? Is the feedback loop, is the React going to go back to the edge? Or is it something that you want to consume internally? What's a tie into revenue for this? What kind of scale are you looking for? Those things start to become significant. So the first step is to say, yes, we need to handle data from things at a very, very rapid rate. But then the next most important question is, what do you plan to do with it? And that's where many companies currently have a disconnect, because they're not really clear about that. So to have a clear line from what you want to do with it, then it can get distilled into, here's how you do it. And let's get this going. Because once that clarity exists, these things can get implemented very, very fast. Yeah. Well, I think, I think, yeah, that's fantastic. So that, that was what I was looking for. And I think, I think to the, uh, to, to the point of clarity, right, that's what I liked so much about your initial model was it was, it was very simple. And then it creates a, uh, like, oh, this is what I'm doing. So sort of realization, like why I would want to go about doing that. So, so thanks for being on to go over that. I think it was helpful. And I'll, um, I'll, uh, in the show notes and I'll probably make it the album art for this episode, if I can, uh, wangle that out, but, uh, I'll, I'll put the, those, those slides I was talking about that have the, uh, the loop in it. So it, but, but before I

start to wrap up, is there any, where should people to go to, uh, to find you, uh, well, not to find you in the real world, but online to, you know, engage with your brand as it were, talk with you more, whatever you, you got a Twitter account or just like LinkedIn where you set up. Uh, yeah, you can find me at, on Twitter. It's Saurabh Gupta SG, all one word. Um, or, you know, just, I'm in the New York office. So just, uh, you know, send in, send in, uh, uh, a request. I suppose my email is as Gupta at Pivotal.io. So just, just write in, um, and, uh, any, any other ways, like any time you contact Pivotal, if you just say, Hey, I want to talk to this guy, I'm sure they'll figure out a way to get it to me. That's right. Go, go through our concierge program. Yeah. You know, I, I have actually been at the doorstep of the New York Pivotal office, but never inside yet. So, so one day I'll go inside. I'm sure it's great. Yeah. And get me some, uh, get me some Sriracha cashews. I gotta, gotta check that out. Well, before we wrap up, I just wanted to mention, uh, that we, uh, we at Pivotal, we're, we're, we're doing our, our cloud native roadshows again, uh, throughout the rest of the year, if I remember. And they're, they're in Dallas, Cincinnati, St. Louis, Hartford, Denver, New York, and Los Angeles. And, uh, they're all in November. So, uh, my colleagues are going to be getting lots of miles to end out their year traveling around. But the, these are, these are basically, I wouldn't call them full day. I think they wrap up around like two or 3 PM, but you get, you get a breakfast and a lunch and you get to see people like, uh, Casey West and Kenny and other people. Uh, I think, I think Fred shows up there to talk about data and they just go over, uh, in general, like what Pivotal is up to and some, some fun zany things. So I'll put a link to it in the show notes. Or if you just search for Pivotal cloud native roadshow, you can no doubt find that, but, uh, you should come check those out if you're, you're in one of those cities. And also if you're very timely and listening to this, I'm going to be up in Kansas

city, uh, on the first and second, which is tomorrow and the day after tomorrow, uh, at a miniature roadshow and also to DevOps meetup. And then if you're, uh, in Omaha, uh, enjoying a steak or talking with your buddy, Warren Buffett on November 16th, I'll be up there doing the, uh, the same thing as well. So there's some exciting, uh, speaking of engaging with brands, you can come engage with us all throughout, uh, the North American area. So with that, uh, thanks for listening as always, if you've enjoyed this episode, or you just want to do something nice, regardless of your sentiment about this episode, you go to the iTunes store and leave us a review or a rating. And the best thing is always to just, uh, recommend this podcast to other people to subscribe to and listen, or just tell us that you like it and then we'll be happy. And the rest of our weeks will be, uh, will be great. And with that, we'll see everyone next time. Bye-bye. Well, we're, uh, clearing out our throats, apparently. Let me get a sip of coffee here. Good call. Because when you want to clear your throat, there's nothing like a pipe and hot cup of black coffee to really clear it out, right? Yeah, well, I'll just swallow some cinnamon right now, then. Second choice. Cinnamon. Did I ever tell you the story that I remember reading maybe in Herodotus, Herodotus, where, uh, the people who controlled the cinnamon trade claimed, uh, one of the reasons it was so expensive was because the cinnamon only grew way up on a mountain guarded by like harpies or dragons or something. And so it was very dangerous. So they would have to take rocks and throw it up at the cinnamon trees. And so the cinnamon trees would break and fall down the side of the cliff. So, you know, ergo, uh, you know, Henry the first had to pay a lot for cinnamon or something. I don't know. That, that seems like a lie by made up by big cinnamon.

Tanzu Talk · 🍕 Get more pizza.