AIOps in the Modern IT Environment – Thomas LaRock, SolarWinds

This is texturing TV. Hi everyone. Welcome back to Tech strong TV the today our next guest and Today's Show is Thomas LaRock.

Thomas is with solarwinds and he'll tell you all about that. But let me introduce some Thomas. Welcome to Tech strong TV.

Hi. Thanks for having me. It's our pleasure to have you on Tom.

You prefer Tom or Thomas. Oh, whatever you think I should answer to Thomas is fine. We'll do Thomas.

All right Thomas. Why don't we maybe start with a little bit of your background? Sure, I currently work for solar winds.

But before that previous and what seems like a lifetime ago, I was a production database administrator for financial services company here in the eastern part of the US. Okay, and how long you would throw it about 13 years now? Oh that is forever.

Congratulations. You know what honestly. I mean, look, I'm in Tech 30 years security 20 almost 25 years.

It's rare today that we see people who have that kind of staying power. Right? And what's nice is you are probably a big Keeper of the tribal knowledge there as well because a little bit I'd like to thank yeah.

I came through an acquisition of a company where we specialize in database performance monitoring software, which is now a crucial part of the portfolio for solowans sure and actually so what is your title these days it's solarwinds. My title is head geek that is another word for senior technical product marketing. Very cool.

I like to headgeek stuff. Yeah now I guess you aspire to be Chief geek. About actually I am the chief geek or the you are the key geek because I lead the team of headgeeks.

Yes. Okay, so you are chief had geek that chg for those who are keeping score at home. Excellent Matt.

So Thomas we're gonna talk today a little bit about AI option observability, which is look I'm getting ready to head out to Amsterdam in about two two and a half weeks. No three weeks and for kubecon Claude nativecon, and I expect observability and AI Ops to be You know one of the top three things there. But solarwinds has a big play and this as well, let's you know, make us smart Thomas.

Sure. So, you know I said I've been with soloons for 13 years. So I've come in at the time where you said let's just say automation was still kind of rudimentary at the time and these days, you know, like you said you're gonna go to a conference next week in Amsterdam and you're gonna hear AI Ops all that all over the place.

What's funny to me is I'm old enough to remember when AI Ops was just simply called, you know, a rum book or right or a bash file or something that you kicked off to run that 7 AM. Yeah. Everything old is New Again.

The idea of calling AI Ops is I I hate business buzzworthy type marketing things but it does kind of help everybody to communicate as long as you understand what you're saying to each other. So AI Ops is really applying the principles of artificial intelligence or automation to some degree to it operations. That's really what AI Ops kind of means.

I think that's the Gartner definition and it's one I kind of adhere to as well. So just if you've Run and if then else script you have essentially done artificial intelligence programming so you should put AI on your resumes, but you can take the idea of some of that automation. But then if you apply like a say a little bit of math to it, then you get let's say it's not just something two or three times better but exponentially better and in my experience where that came into play was about four or five years ago, we built into one of our database Performance Tools is a DPA we built in the thing called anomaly detection now detecting anomalies in say time series forecasting that's not new that is a decades old technique.

But of course, everything is new today, especially to Hipster developers who just think anything with python is just great. So time series forecasting and the idea that I could say. Oh, I can predict the amount of let's say resource consumption inside of a database.

I know what it should look like tomorrow. And that prediction is off. Mmm.

Why was it off now? You should look into that. And what happens is at solarwinds.

We figured out little ways like that where we get the signal through the noise. I it Ops we're all overloaded with information these days. We've got metrics coming at us from every angle and it's really hard to know which one is the right one to be paying attention to so we've tried to go out of our way and figure out hey, how can we raise something and Define a truly as an anomaly by using some math and figuring out hey, this is the thing for you to pay attention to so that's where you see a lot of things inside the solarwinds portfolio.

It's really focused on making it operations a little more efficient. A grid very excellent. So met first of all, thank you.

right because you're on myself get confused about what AI Ops is and we got to explain it on here. And I know a lot of people out there get confused about what AI Ops is exactly too and what I like what you did in your definition there. You brought it to.

Automating the query if you will, right? But now we also have this thing that people call ml Ops and depending who we talk to some people say ml Ops is AI Ops or AI Ops is ml Ops. Are there people say?

No. Those are two very different things. Right, and I don't know what to believe.

What would so what's your give some you know again make us smart Thomas here. I'll make it. Well you I don't know I'm making you smart.

But you know what? I'll do my best to explain this so okay. Here's the thing.

I artificial intelligence is like the the if you have a Venn diagram, it's the biggest circle. It's the plane itself. Right?

It's artificial intelligence is simply any task that computer can do that a human once did if Alan Turing was sitting right next to me pretty sure he'd agree with that because that's how he defined it and that's the person I'm gonna reference. Right? So if Alan Turing says this is what artificial intelligence means I'm going with it.

Now inside of artificial intelligence is this thing called machine learning you should think of that as statistical models. You don't have to think of it as anything other than that and again nothing new right? We've been having Actuarial science for decades.

This is how insurance companies have set rates for centuries. They She was a little bit of math. So that's what a machine learning model kit will do.

The basic ones would be like a linear regression. You don't have to worry about any of that just understand that machine learning is is a subset of artificial intelligence. So that's why somebody might tell you.

Hey, it's the same thing. Well sure there's a Venn diagram in play here. Right?

Sometimes it's the same thing. Sometimes they're very different things. So you can't you just have to think of it that ml is more about building a Model A model that's going to do prescriptive descriptive perspective.

Right? So you got models for different things and you just want to be able to maybe get an answer or figure out what you're supposed to do next. That's all ML and automating a lot of that model, right?

So inside of some of our products we do a little bit of that right we take the data every day. It's automated. We build a prediction then we try to get smarter.

We feed it back into the machine. That's ml Ops. So AI apps ml op They're they're really defined as two different things.

Do they have an intersection? Absolutely. Agree, very cool.

now Let's talk observability and how that plays in there. Okay? So observability another buzzword right?

I'm old enough, you know to remember when observability was sort of just called Full stack monitoring say observability isn't anything new. I know they talk about it coming out of control theory where you infer the health of a system by observing it if you can observe the system and infer the health right that that old old definition, but I think of observability as this as understanding everything all the layers between an end user and the data and Back Again. And the idea of observability is really a lot of it comes out of the Google paper about 15 years ago or so for database reliability engineering where they said.

Hey, we just need to know about concurrency errors latency and throughput as a DBA. I know there's a lot more than that, but the high level if you're into say devops and you're doing these iterative programming and change control you making a slight change then you want to measure things and you apply the idea of observability now, I want to observe what's happening. All right, that's fine.

That's fair. That's one way of achieving your goal. But That solarwinds we do Enterprise infrastructure monitoring right?

No matter where you are Cloud Earth doesn't matter Enterprise infrastructure monitoring. We want to give you that visibility the observability to every layer. So we're talking switches and routers.

We're talking storage. We're talking databases not just the database the metrics and the server but the metrics inside the database engine itself. So we're talking concurrency locking and blocking and deadlocking.

We won't be able to give you all of that. We want to be able to give you the network metrics all the things that database servers can't necessarily give you virtualization. We want to give you it insight into the what's happening in the virtualization layer if you're overloading a VM kernel and that's causing a bottle neck which may look like storage or network, but it really isn't, you know, we want to be able to bubble that up again with anomalies.

We want to be able to give you an idea of that something is different that you should look at right now. Love it. Good.

Good good. And I do because you know what Thomas I'm older than you. You know what I and I so I used to what we call AI outside.

I used to think of that, you know as application management. Yeah, right and and observability was it was sort of full stack, but I mean we've been doing it or trying to do it for as long as I've been in Tech. Yeah, the father's experiment I give is when somebody talks about observability inferring the health of something.

I say. Okay, great. There's a dumpster.

Is it observable? Yeah. Can you infer the health of that dumpster?

Well, is it on fire here? He's now that that yes. I do lead to a toaster fire.

Right and therefore I am but anyway. Let's talk a little bit about solarwinds, right? How did they bring this all together?

Yeah, so I mentioned the anomaly based detection inside of database performance analyzer but we've been building a lot of AI Ops or let's just say Ai and ml into a lot of our tools. The latest one that I think I'm really most excited about is anomaly based alerting. So if you're like me, you're old enough to remember where you would be bombarded with alerts just information overload, like something's gone wrong.

You should look at this. Now you need to look at this thing. So what we've tried to do is make things a little bit smarter again get that signal through all the noise where we're gonna alert you and you can even build these conditions yourself and just sort of say okay only know the find me when there's really something I need to take action upon right alerts should require action.

Everything else is just information that can be logged later and not only base a learning is one thing but one area where we've done this, but in a lot of our products you will find an application you'll find linear. Russian models being used in order to make something a little bit smarter. You'll find math in probably a dozen different areas of our tool at this point, and we're just building more and more every passing release.

excellent Look, I can't do an interview these days without getting into the whole chat GPT ai iterative ai. Okay, how do you see this? Influencing or you know becoming involved here.

So chat gbt right now to me. Is it still just a toy and it can be fun at times? It's useful but it's like any other toy or tool it has a place.

But it's not necessarily the right tool for the job. You're trying to get done and I've knows people who are who say oh, I asked it to write some code. Look what came back and other people will tell you that code isn't going to work and I've done it where I said write me an article on this and what comes back I look at and I say, you know, if a student tried to hand this to me, it wouldn't be acceptable either.

So if you have some domain knowledge, I think chat gbt helps when you ask it something you might get back a result and 75% of it could be unusable. But what the alerts you to one the two things you didn't think of or that you weren't aware like oh, okay. Thanks for the reminder.

So in the way, it's almost an enhanced search, but I'll tell you this last week. This is what I did. I told chat TBT I said, hey.

Uh create a travel itinerary for me where I can visit every Major League ballpark to watch a baseball game this summer minimizing travel time and cost. And I swear it came back with something that mostly works for me like it. Yeah, you can start in San Diego which I love already and then you're gonna go to La and then you're gonna go say and it chart it said here's the game you'll attend.

Here's how you get there. Now certain things, you know Seattle to Phoenix some I guess I would fly or maybe I might be two days worth and I could drive that but it was interesting how it knew where to go get pieces of the information and to give me a basis to start upon and again, it's all how you phrase the question right? So I thought that was extremely useful to get started with something that I would never want to take the time trying to do myself.

Absolutely, and I think look I think. I don't disagree with with a lot of what you said, I think. it's movie toy level to Tool level right But yeah, I think even the fact that you said I told you I told check that GPT.

Well, you didn't tell it anything you typed in something right? Let's be clear but the days coming where you going to tell it and it's going to spit it out for you. It's probably It's not that far off, you know, we're in four.

Maybe that's a four point five thing rather than texting, you know, just put a seat voice to synthesize her on it and whatever and now you just talk to it, but that that being said, yeah for that kind of stuff it is it's it's really good. I was talking to someone the other day who basically types in, you know plain text and it kicks out sequel queries. Sure.

uh, you know, so there's a lot of things I I don't know. You know how it's going to affect sort of AI Ops and observability. It may just really change.

It depends look data. Crap data in is crap date or out number one. Yes.

So, how can you get your data into it? I and then you know and asking the right question is is a big piece of it, but I do think it's gonna have a role in here if we just Got to see how that plays out. Anyway.

Hey man, we're overtime Thomas. I want to thank you for coming on and as I said making a smart on a few things, some of us are hopeless case. I love you here.

Appreciate you having me back. It was great. It's always a pleasure to have you on don't stay away so long man.

All right, are you gonna be at RSA? I will not. I will not my colleagues be there stop by the solarwinds booth.

Say hi to Crystal Taylor. She'll be the head geek in the booth at RSA. But not the chief had geek now, but alrighty Thomas Thanks a lot, man.

Hey, I'm serious. God it's a pleasure always have you and you know, what for people want to get more information on solo wins. com, right?

Yes, sir. Alrighty, we're gonna take a break on tech strug TV. We'll be right back.