The Art of Network Engineering

Ep 130 - Exploring the Influence of AI on Network Engineering

A.J., Andy, Dan, Tim, and Lexie Episode 130

Send us a text

Imagine, if you will, a world where AI streamlines your network configuration changes, manages your network and even troubleshoots for you. Sounds like a far-off dream, right? Well, the future is here and we're discussing exactly that. We weigh the pros and cons of using AI in network management, touching on issues like data accuracy and false positives. But it's not all gloom and doom as we explore the positive impacts conversational AI can have on the networking industry.

Finally, we throw light on how AI can revolutionize network security and enterprise. Picture IT administrators making informed decisions about upgrades and patches swiftly and accurately, thanks to the assistance of AI. We consider the implications of using public data lakes and protecting intellectual property. We then delve into the exciting potential of AI for network optimization, self-healing, and even ordering new equipment. Trust us,  this is an episode you don't want to miss.

A big thank you to our sponsor: Unimus! Interested in a device agnostic NCM software to streamline and automation configuration management for your entire network? Check out Unimus! https://unimus.net/

Find everything AONE right here: https://linktr.ee/artofneteng

Speaker 1:

This is the Art of Network Engineering podcast. In this podcast we'll explore tools, technologies and technology keeping. We aim to bring you information that will expand your skill sense and toolbox and share the stories of fellow network engineers. Welcome to the Art of Network Engineering. I am AJ Murray at no Blinky Blinky on Twitter and I am joined this evening. Flanked, flanked I'm going to use flanked and flanked this evening by by Dan at Howdy packet. Dan, I feel like it's been a while since we've recorded together, so I know it's been a while since I've recorded. I've done a crap ton of editing behind the scenes, so I'm excited to be back behind the mic. Dan, it's good to see you how you doing.

Speaker 2:

It's good to see to AJ back. When I was a teenager, I started getting into PC gaming right back in, probably like my 13 somewhere around there.

Speaker 1:

As we all did.

Speaker 2:

Yeah, you know, I kind of stepped away from you know the, the Nintendo 64, PlayStation, xboxes, and I stepped into the PC world, right, and I haven't gone back since then. Right, it's been pretty much online multiplayer games on PC since then. Well, this past week I've been, I've been wanting to get my boy into video gaming, right, he's, he's, he's, he's four years old, he just turned four. But PC, just it's not really a good fit, because one I would want him to be on a separate computer so that way I can, I can do my stuff on my computer and he can do whatever on his. So I was like you know what, probably need to go to the console route, right, and so I bought him a switch, the Nintendo switch, and we got like Luigi's mansion and some Mario Kart, you know, kind of some of the classics, right, yeah, and, and it's been interesting watching him learn that, well, also, I kind of fell back into my old childhood a little bit.

Speaker 2:

Because, I didn't play, you know, story mode single player games, right, like for since my teens, right, I've been. Once I got online and I started playing multiplayer games. That was. That was the rap right there. Well, I've been playing the latest Zelda and it's just crazy because, like, I'm really enjoying it, there's so much in depth to that story and and like the playing of it, it's just been a weird, weird, I don't know vibe that I've been having here lately. I've been actually wanting to play single player story mode type game and I don't know that's that's been. That was a very long intro.

Speaker 3:

I'm doing hang on so so, but that's where I'm at right now. You wanted to get your, your son, into playing games with you and then you started getting single player games and tell him to piss off.

Speaker 2:

Yeah, no no, no, no, no. After he goes to bed.

Speaker 4:

I've been playing a couple of those.

Speaker 2:

Yeah, sorry about that, that was a really no, no, no.

Speaker 1:

So I saw.

Speaker 4:

I saw the new.

Speaker 1:

The new Zelda game is out and I have yet to pick it up because the last Zelda game that that came out.

Speaker 1:

I picked it up and I was like oh yeah hard, like I played that game hours a day when I should have been like doing anything else, and so I will get it. I just when I'm in a, when I'm in a better place in my life, I will get it, and I don't mean to say that I'm in a bad place, it's just I have a lot of other priorities right now that I really need to take care of. So I feel like if I pick up that game like all that, that shit's just going to go like out the window.

Speaker 2:

Yeah, it's, it's wild. I was actually surprised because, like the last Zelda game that I played was Majora's Mask, right.

Speaker 1:

I don't know if you remember that one, that that was in 64.

Speaker 2:

You know, I can't remember what year that was, like 98 probably, or something like that Maybe earlier, but in I do. I played hours on that game and it had nowhere near like in depth. What you can do in the game as the tears tears of the kingdom, I think, is what is called the new one, and it's just, I'm just blown away. I've been having a blast with it, but anyways, anyway, he digresses.

Speaker 1:

I was gonna say I digress, I didn't digress.

Speaker 2:

Yeah, I'm the one who digress.

Speaker 1:

Tim Bertino. At Tim Bertino on Twitter. How are you doing, Tim?

Speaker 3:

It's the time of the year we got to mow the grass oh yeah, Socks.

Speaker 4:

But I feel, bad for you.

Speaker 1:

I have a teenager.

Speaker 4:

I have the law and you sure do like three or four years.

Speaker 1:

But I'll my teenagers turning era. He is 18 and he may fly the coup later this year, so I might have to go back to I was gonna say I wasn't gonna bring it up. It's like and he got out of there, yeah.

Speaker 3:

You bet, I'll continue in the vein of video games, because you brought up some nostalgia. I mean, it was like tradition back in the day. It seems like for our generation. I have no idea what teenagers these days you know how they binge games and whatnot, but you'd hit like a holiday, your birthday, christmas, whatever. You'd get a new game and it was like for 72 hours straight, oh yeah, you'd get through it and then it'd be on to the next one, or you. Then you'd maybe go back outside, maybe and do something else. But yeah, that was I was big Dan. You brought up the N64. I was big on that and gold.

Speaker 3:

Yeah, yeah rogue squadron of Star. Wars was probably one of my favorites flying around, and that was Star Fox 64. Yeah, man.

Speaker 2:

Yeah, and so you can actually get the online what's? It's like Nintendo online account, blah, blah, blah, something other multiplayer, I don't know. Yada, yada, yada, but you can play Star Fox 64.

Speaker 1:

You can download some old N64 games like the old man, yeah, yeah.

Speaker 2:

And you can actually play it on the, on the switch, so that I actually did that. I started playing Majora's Mask game because I was like that's my childhood. But yeah, to your point, tim I, it was Christmas of 96. I was six years old. I got the Nintendo 64 and Mario 64 and Star Fox 64. Those are the two games that I got with it. I are the morning, I mean, and I went straight to my room and I played. I played Mario the entire night. Yep, I didn't come out and it was Christmas late.

Speaker 3:

I'll see you all next year. Yeah, what's going on with you?

Speaker 1:

AJ, oh, man, a lot. You're right, it is spring. Spring has finally sprung here up in Vermont and we have a ton of projects going on outside. So at the end of last summer, early fall, we got a driveway put in. I've had a stone driveway for the longest time since we moved in and and every time I would snow blow it would chuck stones all over my front lawn and then, after the snow melted, I'd have to like rake the lawn to get all the stones back out, put them back in my driveway.

Speaker 3:

We don't have that problem. You've mentioned that before. So were your snowblower blades just all like freaking mangled, or did they all go pretty well?

Speaker 1:

It's got like feet on it and so I would like jack up the feet. So it was okay. Barely like you know, tickle the stones. So anyway, we got the driveway put in but the driveway added some height to. You know, the asphalt added some height to our driveway so we had to get a bunch of dirt to like fill in around it. So I've been trying to cultivate that and grow some grass. Okay, so that's been my project. What?

Speaker 2:

do you all grow up in Vermont? Is it because, like here, we do mostly like fescue and that kind of stuff?

Speaker 1:

I couldn't tell you the variety of grass, Dan. It's just some grass seed that I threw down and AJ, this is like like I put hold on.

Speaker 3:

Wow, that reaction from Dan was like those thems is some fighting words. You don't know the name of your grass.

Speaker 2:

Look, look. Okay, there's different types of dads, right there's. You got some better the barbecue dads, which you know, that's Andy, but you got some yard dads. You got beer dads.

Speaker 1:

I prefer to think of myself as like a little bit of everything, right. Okay, your jack of all trades, master of nothing. I don't delve that deep into anything.

Speaker 3:

I don't think Alright. So, dan, are you out there in the, in the white sketches, in the jean shorts? Oh, it, not yet.

Speaker 2:

I'm getting to that face. Like like the high ankle, high socks. Oh yeah, it's, it's gonna happen Right, okay there's. There's gonna be a moment and it's gonna be like it's just gonna happen overnight. I'm gonna cut the jeans right when they're gonna be.

Speaker 1:

Jorts, that's what we call it down here, jorts, yep yeah.

Speaker 2:

And and yeah, I definitely want to get the white sketchers or new balance new balance Cheese.

Speaker 3:

I even screwed it up new balance.

Speaker 2:

Yeah, you already got to figure it out, Dude.

Speaker 3:

I'm planning. I'm planning. You said it was gonna happen overnight. Did this happen last night?

Speaker 1:

Yeah, it's gonna be like a caterpillar you know how they?

Speaker 2:

yeah, I just got to turn it like cocoon himself up there and that upstairs.

Speaker 1:

Yeah, he's gonna emerge one day with Jorts and his shoes tucked in, tucked in t shirt. You know, of course, absolutely yeah.

Speaker 2:

Yeah, you know it's coming.

Speaker 1:

Anyway, lots of, lots of stuff going on here around the Murray home. I was telling you guys, before we started recording the show, my, my modem died earlier this week. That replace today. My router died and I was kind of scrambling because it was in the middle of the work day. It happened to be the one day that my wife decided she was gonna work from home, and so instead of running out and buying a new router, I was like, oh, I have a lab here full of gear and I took one of my ISR 29 hundreds and I repurposed that for my home router and so hey, back on it's only lab until it's production.

Speaker 3:

That's right.

Speaker 1:

So it's a. It's a nice temporary fix until I can figure out what's going on with my other my other gear.

Speaker 2:

But that's good to hear that a 29 hundreds rock, and what'd you say your downloads were?

Speaker 1:

I'll get like eight or 900 down.

Speaker 2:

That's a nice.

Speaker 1:

Yeah, that's good for a lab. Yeah, yeah, not bad at all, especially because it was like a 2901 too.

Speaker 2:

I really wasn't expecting to get the full.

Speaker 1:

Speedback.

Speaker 2:

What year was the 2900 like released? I mean, that's been a while right.

Speaker 1:

Yeah, it's been a moment. I want to say those were like 2011.

Speaker 2:

Hmm.

Speaker 1:

I don't remember.

Speaker 2:

Yeah.

Speaker 1:

Really don't remember, but that's not why we're here Tonight. Uh, we are gonna dive into the topic of AI and networking. Um, you know, tim and I we were at NFD earlier this year and we saw a lot of products that are integrating AI. We just returned from uh, dallas, texas, where we were at the juniper enterprise analyst and influencer event and we saw a whole lot more AI in in what they're coming out with later this year in their networking products and, uh, there's also been a lot of discussion around utilizing AI to build networks like Chat, gpt and some other things out there. So, uh, where do you guys want to? Where do you guys want to start with this one?

Speaker 2:

I would like to start with uh, network management. Okay, all right. What? How could we leverage AI to help us with automating, let's say, task and normal Configuration changes? You know the the day in, day out, just normal stuff that you're you're configuring right. How could we leverage AI for, for some of that?

Speaker 3:

So I kind of want to frame that up a little bit, because the the biggest networking term, as far as in my mind, as far as automation around configuration, especially around the data center, for the last few years has been intent based networking. Right, the, the, the thought that you have this either gooey or api level interface to where you give your high level intent this is what I want the state of my network to look like and the Automation system, the controller, however, that's delivered, abstracts what it's actually doing, the actual underlying network configurations, away from the administrator, from from having to do so. That's kind of where we've gone, uh in the industry lately. So I don't know if, if AI is just a spin-off on that or takes that the next step to where we're leveraging AI to define our intent. I I don't know, aj, what's your thought on that?

Speaker 1:

When I think of controllers, I don't know. I mean, until recently I haven't really seen a whole lot of like AI in the controller. Right, the controller is taking a known, good templated design and allowing you to plug in some variables and then deploying that, that design. I don't know that it's necessarily AI taking business requirements or or user input and designing or deploying Something like that. Right it's. I think it's, you know, a cookie cutter network to. To boil it down, right, it's not. It's not that simple, right? Especially when you're talking about something with underlays and overlays, like vxlan and lisp and and all that good stuff, are you're in the data center or the or you know the enterprise.

Speaker 1:

But but AI, you know, we saw a lot with the, the juniper mist offering right, where, if you're having issues, it could analyze, detect and propose potential solutions. It wouldn't implement the solution without an operator selecting the proposed solution or approving the solution get pushed. And then we're seeing other things like truly proactive Notifications that, hey, we're seeing behavior that looks like an sfp is about to fail. We're going to get you a replacement sfp so you have time to do a maintenance window and swap it out Before it actually fails.

Speaker 3:

So you, you just jumped into what I think is the biggest use case for at least AI in the near future in networking. The three of us have been around enterprise networking for a fair amount of time now and to me the the biggest pain point of Operating managing networks is troubleshooting.

Speaker 3:

Yeah, yeah finding issues Uh, as quickly as possible, hopefully before a user calls in, which rarely ever happens. I mean, there's some really good basic monitoring solutions that can do uh alerting and ticket generation and that kind of stuff that I've been exposed to. But what I'm talking about is Getting out in front of those issues and, if somebody does happen to call in with an issue, having as much data in front of me that's in a readable format and maybe even points me toward the next direction, to your point, you were just talking a j like the sfp thing, and then also suggesting what's my next step and giving me maybe someday the option to not even Ask me what to do next, but but just implement that automated remediation.

Speaker 1:

Instead, instead of asking what do you want to do next? It's here's what. Yeah, hey, I did this.

Speaker 3:

Um, so what one one that comes to mind. I know we had multiple vendors that presented to us at nfd earlier this year and uh, about things like this, the automated remediation, or at least log gathering, to point you in the right direction. And that selector ai was one, uh, that that really stands out of somebody who's Uh a company that's leveraging ai today to gather those different logs and do that correlation. And that's where I think the difficult thing is is you have all of these different systems, all of this different type, all these different types of data. It's you have to have an ai system that can compare those apples and oranges and be able to give you meaningful output. And I to me, I think that's the, that's the biggest challenge, but that's that's also, I think the biggest reward right now is network automating, using artificial intelligence to automate those troubleshooting network operations types of tasks, right.

Speaker 2:

Yeah and like. So one thing you were kind of saying is challenges with that is like, what about like data accuracy, right, like if you, if you're fat fingering things, you know the is ai going to be able to help you to correct that? Or, um, like, what about exceptions?

Speaker 3:

You know, how does ai handle exceptions, that kind of stuff and can you get your ai to Uh like, talk back to you in certain ways, like you do? So you just said, dan, am I not, am I Doing something dumb? And your ai just comes back. Hey, dumbass, you know, have that, because that's a thing too right, that conversational ai, that's something with leveraging like chatbots to actually, um, communicate uh, aj was mentioned in juniper, and and marvis and being able to start having that Conversational based network John capo bianco has been talking about that too. Oh yeah, of using Automation and ai to have conversations with your network. It's, it's far out to me, man, but it's coming.

Speaker 2:

Yeah in. Oh yeah, go ahead AJ.

Speaker 1:

The, the conversational thing is interesting to me because even today, if you have, like, a home assistant and you ask a question but then you want to ask a follow-up question, it it doesn't carry that context over, right. So so using something like chat, gpt, where Within that, within a conversation, like it remembers stuff that you you asked about or information that you provided early on, earlier on in the conversation To, to keep that context, you know, consistent or alive, rather than having to, you know, continually restate things over and over again, which I find very interesting. And so In in the network troubleshooting context, I think that's really important because if we're talking about a single incident to, to be able to Maintain you know what the problem state is, and then try all these different troubleshooting avenues and still remember what, what's the original Goal here, what's the broken thing we're trying to fix, like that's, that's going to be huge and making that like a really viable reality.

Speaker 2:

Yeah, but also one thing to add to that too is like uh, you know, because I I want to have this conversation where we talk about what are some challenges to.

Speaker 2:

I'm not going to be like one of those dayticker jibs kind of people and, like you know, AI is going to replace me and, and because I mean it may, it might do that, right, but I don't have any control over that, so I'm going to use that as a tool. Uh, you know, in my favor, um, until that time comes, if it does. But there's, I feel like there's going to be some challenges with it because, uh, like, where do you draw the line between something that AI needs to do versus a human needs to make the decision for it, right?

Speaker 3:

Yeah, that's 100% correct and that's where I go back to. To me, the biggest place that that makes sense is in troubleshooting and operations. I'm still not quite there Now. Granted, I'm not a heavy cloud person and I'm sure it's probably much more prevalent in that. So I'm thinking enterprise networks, enterprise data centers. I totally see the use case for AI in troubleshooting and day to day operations like that. Where I struggle with it is in actual building out of networks, like I'm finally coming around to the whole what we were just talking about a little bit ago the network state, the intent based networking, setting intent it a GUI or an API, and really at abstracting all the underlying configuration away from you, so you don't necessarily have to know what all of that is. You're not configuring the underlays or the overlays, you're setting your intent, essentially walking away and hoping it works. Yeah, I just I don't see the yet, the actual AI in building networks. Now, dan, you just did the. You took our jobs.

Speaker 1:

Yeah, so I want to call out and you didn't disappoint.

Speaker 3:

I do want to call out a recent episode. It was episode 42 of the heavy strategy podcast of packet pushers and it was actually called is chat GPT coming for your job?

Speaker 4:

And in the description.

Speaker 3:

It was great In the description. Its short answer is yes, long answer is long Chat. Gpt is useful for demonstrating potential and getting more funding for AI. And then it posed a question of how can it partner with humans to increase productivity and improve technology experiences, and that's exactly the question you were posing, dan.

Speaker 2:

Yeah, well, and one thing I want to kind of go back on that you were just talking about is how, how do we leverage AI in intent based networking? Because I'm right there with you, I'm just I'm struggling to understand if we have a template that to build out, you know, a data center, I get the automation side of it right, like but that's not AI, right, right, right.

Speaker 2:

That's what I'm at with it, yeah it's, I'm giving it a template and then it's going in configuring that template, you know, to make the network configured as intended, right, but how does we leverage AI? And here's where I think we might, where we might be able to leverage, that is we. You know, in our in our recent Juniper discussion that we had, I brought up the question of okay, the fabric is going to see the traffic. Right, I mean, it has to. So if it knows that, you know, application A is trying to talk to application B, what if there was a way for AI to be like hey, we noticed that this, this application is trying to talk to this application. Do you want to allow that? That policy, and then either you say yes or you say no, like we want to go that further. You know, I feel like that could be a use case for AI in a template type network.

Speaker 3:

So I yeah, that's. That's an awesome point, dan, because I wasn't originally think I don't know why I wasn't, but I wasn't originally thinking the like the data center security aspect of it, because that is so difficult right now is we have these intent based networking solutions in the data center that can do that level of micro segmentation, which is great, but you better know how your applications are flowing to be able to you stuff. At the end of the day, you still have to write those policies, yeah, and that is so difficult. So, to your point, to start leveraging AI solutions like that that all they need is taps or spans or whatever, a way to get that data or, like you said, an integrated fabric, a way to to mirror that data, to then be able to leverage AI to suggest policies based on baseline workflows, that's yeah, that's a fantastic point.

Speaker 2:

Because even even though it's intent based, like kind of to my point in the in the juniper episode, not not, I don't think everybody knows what their actual intent. Is right, like it's a learning phase to some extent.

Speaker 3:

Yeah, I don't know what the hell I'm doing?

Speaker 2:

Yeah, me neither. I mean, I don't know what.

Speaker 3:

I mean until it works, and then yeah, that's what I meant to do.

Speaker 1:

I think if we're talking about, like AI, you know, potentially replacing jobs, I'm thinking about the analyst, security analyst, like any sort of like work analysts, like people who have to, like pour through logs and make heads or tails of you know what's something worth chasing down versus oh, that's normal kind of thing like that to me seems like something you could train an AI to look at traffic and then notify me, like when you see something that's truly an anomalous right, like when there's actually a problem.

Speaker 3:

Yeah, yeah, that's a good point, the whole, you know, leveraging AI to in the the same, the, the actual security log aggregation and in correlation yeah, definitely that that could be very beneficial there.

Speaker 1:

Those tools are great and they do a lot of like really powerful things, but they need somebody there to look at that information. And from my perspective and in my own experience, I've been held by, you know, certain security compliance that says you have to have these monitoring systems and these tools. But I barely had the budget to buy and I certainly don't have the staff to operate it, and so it just became this tool that yeah, sure, I have a syslog collector, but I don't have somebody that can stare at that thing all day and we only ever touch it when we're doing a troubleshooting.

Speaker 1:

You know adventure.

Speaker 2:

Well, and then also it's like and now and I believe some, some of these SIM solutions have this capability, but like, let's say, you have to be. You know your company needs PCI compliance, right, you know, if you had an AI, that was, you know crawling the logs and all that in there like, hey, this, this, whatever just happened here, is not normally PCI compliance. Do you want to do something about it? I think I don't know if there's any systems out there that can do that specifically, but I do feel pretty confident that there's systems out there that will let you know that that's out of compliance. But maybe the AI portion would be do you want to do something about it? You hit yes and it goes and configures kind of back to that intent based you know networking where it let you know that you got out of posture and you're saying or out of compliance in your posture and you're saying, hey, I want you to go in and fix that and then let it do its thing and you know, configure the network properly.

Speaker 3:

Yeah, like in that scenario you're you're stating my intent is PCI compliance. You tell me if I'm out of compliance and in where and what I need to do to fix that.

Speaker 2:

Yeah, Right, and then let the AI figure out what it needs to do on the back end, with automation to make it happen For sure, see, I keep going as soon, as soon as you mentioned it a few minutes ago, dan.

Speaker 3:

I keep going back to the security use cases. Yeah, because security is all about having updated, meaningful data to be able to act upon. So I know there's all different types of threat data feeds that you can subscribe to. So to be able to ingest all those different feeds and then be able to correlate against them to find out what's the best thing to do in the environment to potentially prevent vulnerabilities from being exploited oh yeah, I don't see why I could not be make a play in that kind of space.

Speaker 1:

Absolutely. I mean, you get all the CVs that get published and you know there might be one that applies to a piece of software that you're running, but you're not running the service that affects it.

Speaker 3:

Right, that's a big pain point. Yeah, that's a big pain point right now.

Speaker 1:

I understand like am I affected or am I not?

Speaker 3:

Right, because I mean, any vendor will publish the vulnerabilities based off of CVE, but a lot of times it's going to be you'll get it, it'll show the affected version, and then you have to read through to decide, because they'll show you in the fine print Okay, yes, it's this version, but you're only affected if you're running this feature and have it configured in this specific way and, I think, having those, those monitoring systems.

Speaker 3:

Right now, I'm going to say this I don't have any data to back it up, but I'm going to say it's probably more often than not going to be yes, you're affected by this because you match this version. And then you need to dig in and see if you're really affected, and then you need to prove that to the InfoSec team, so to be able to have something that's able to understand not only the intent but also the state of your network and then say, okay, I get this alert. Okay, yes, it matches this version. However, this function of this feature is not enabled, so this is a non-event. And not only be able to have that visibility, but to be able to report on that and show that, nope, we're not affected. And this is why, without having to spend an hour trying to figure it out yourself.

Speaker 2:

Yeah, and actually let's pivot a little bit there, because you brought up I didn't even think about this at first. But what about, like network provisioning, right, like let's say you want or maybe not even network provisioning, but if let's say you want to convert, right, you want to go from one vendor to another vendor. Actually, aj, I think you were just talking about this what if there was a way that AI could figure out okay, this is what the config is like on you know this vendor, what does the config need to be to match whatever your features and your settings and protocols and all that stuff? What does it need to do to basically translate that into the new vendor? Or here's another one is like how many times have you guys had to go through like data sheets and stuff just to see if upgrading to this new version is going to wreck your platform or not? Yeah, what if AI could tell you that, like, it can dig through all those data sheets or white papers where everyone's home.

Speaker 2:

Oh, bug scrubbing too. Yeah, bug scrubbing, exactly. And what if it could let you know like, hey, you know, if you upgrade to this you're going to run into this issue or this issue, or it can come back and say nope, you're good, this is a safe move for you guys. And because I've heard horror stories about people doing this and they just missed a small section in these data sheets. What do they call data sheets, white papers?

Speaker 3:

Yeah, data sheets that specify the different specifications of the gear.

Speaker 2:

Yeah, yeah, and I've just heard horror stories, though, like they're down for hours because it just bricked like an X5K or something like that. You know what, if, what if AI was there to help guide or, you know, ease your worry or concern about doing those kind of those major upgrades like on core switches or stuff like that?

Speaker 1:

right, yeah, because I mean, the last thing that you want to do is upgrade to avoid one CVE, only to find out the version you went to has like 10.

Speaker 2:

Yeah, yeah, yeah. Bringing it back to security. Yeah, because that is a very.

Speaker 3:

That is a very manual process. I mean, that's something that I try to go through is I'll go to a vendor website. I'll try to find the you know, the starred release in the maintenance version. I'm not wanting bleeding edge, I just want the long live maintenance release for the most part to get the bug fixes Figure that out. Go through.

Speaker 3:

Like Dan, you both just said you want to make sure you're patching a security vulnerability or a potential security vulnerability, but you want to make sure that that patched version doesn't contain its own. So you're often you might be looking at the security vulnerabilities of that vendor in those versions and then you may pivot to the operational bugs list. If that's on, you know a different platform or a different list. So it would be cool to be able to set that intent that we're talking about as far as software and say, basically, just set the intent of I want the most stable long live maintenance release and the AI system already has all that data from your network. It knows how all of your devices are configured, what features you're using and you can tell it. I want the most stable version without you know the least amount of operational and security vulnerability bugs, and it's able to figure it out and spit out. Okay, we believe you should go to this version and, by the way, we've just pushed it in stage. You're done on all your devices.

Speaker 1:

It sounds like a dream? Yeah, it does. It's like a utopia. It's like an IT administrator utopia.

Speaker 2:

The only concern I have with AI in that regard is like what about false positives? Like how trusting can we be with this?

Speaker 1:

Like AI, like any other computer issues zeros and ones. Right Like it's, it either matches the condition or it doesn't. Right and so and so when it thinks it matches it doesn't know like, well, that's a false positive kind of thing, right, like it's a yes but it's a no.

Speaker 2:

And like one thing I've seen is like using AI in camera systems. Right, they will kind of they'll put a percentage on how sure they think they are on something. I wonder if there's a way that they could integrate that into like decision making and stuff like that. Like if your AI comes back and it says, hey, I feel 89% sure that you upgrading to this new code is not going to break anything, right, is that enough for the business side or the engineering side to be like all?

Speaker 3:

right, we'll send it. On that. You know I was going to say, dan, I need to know what percentage is your send it factor?

Speaker 2:

Yeah, and see I would. I would want to train my AI to be like look like Dan. Do you want me to full send or do you may just kind of send it?

Speaker 3:

Kind of send yeah, yeah, 45% certainty close enough. Yeah Full send.

Speaker 1:

I think that would require additional data, though. Like to train the AI, right, like it would need to know. Like okay, based on your hardware platform, based on the software version you're coming from and the software version you're going to, 89% of the time the it works every time. Yeah, exactly.

Speaker 3:

So that's where I wonder if and I think this is what's interesting with trying to leverage, like cloud data lakes and anonymized information potentially with other customers of the same vendor, because I think that's where you could leverage that, that type of information to say, okay, other customers are doing this, this is the success level they have, and that's where I think AI could be huge, with technical assistance center tech teams of vendor support is that they already have this data. Oh, yeah.

Speaker 3:

And they're already able to parse it and go through it and try to find. You know how many times has this happened to other customers and how did we fix it? So I would think that AI has a big use case for that as well.

Speaker 2:

Tim, I was about to say the exact same thing. Are we going to get to an point in time where, when we set up, you know, network devices, is there going to be an option just like on, like when you set up your Windows machine hey, can we send data back to Microsoft?

Speaker 2:

You know, is there going to be a time where we're going to basically be Selecting a checkbox that says yes, use my data In in. Is that dangerous? You know, like I could definitely see government not wanting to do that, and I would think some larger customers probably wouldn't want to do that either, but it if it's all for to help AI learn what's best, what's not best and make good decisions to better Everyone's networks. You know, I don't know. Maybe maybe you should. We should be sending that data.

Speaker 1:

So when I look at some of this AI stuff, I see it as like a wizard, right Like you, and you're installing a piece of software.

Speaker 4:

You're a wizard Harry.

Speaker 1:

But. But right like, but I'm a VM, we're in a Windows guy for a long time, I. And a long time ago, when you wanted to stand up something you had to like Install Windows first, patch Windows, put on, you know your AV and and all that good stuff. And then you got to the application and you put in the application and you configure the application and you configure the you know Windows firewall or third-party AV software, slash firewall that you're putting on there to allow that communication to happen. And then you fast forward a few years and you know VM, where it gives you vCenter VM that you can just deploy. It'll ask you like six questions and it takes care of everything for you, like that's, that's taking what used to take hours upon hours worth of work and then compacting it down to, like you know 10 or 15 minutes.

Speaker 1:

Right Like, why you're waiting for that stuff to deploy. And so all of this network automation is doing the same thing for network engineers. Right Like, if I have to stand up a network from scratch because it's Greenfield, or for doing like an upgrade or something like that, why not have a wizard? Right Like, give it some input, give it some variables and let it go and let and let me focus on Tasks that need me to focus on them, rather than doing some of this like mundane Common tasks stuff well, and to that point, a little bit of the, the deployment side of it.

Speaker 2:

We haven't really talked too much about. I guess that. But like what, if we're doing like brownfield deployments Is introducing something, a change, or you know new hardware, it can AI, crawl it and figure out hey, you know, this is gonna cause an outage because of this right here, right, oh, yeah, you know what if it could give you hey, this is low risk, we don't see any issues, you know that kind of thing. Uh, I, you know, especially with your role a j, I think that would be quite amazing, I mean well, you talked about going from one vendor to another.

Speaker 1:

That is something that we constantly face, like I see it in work chats all the time like hey, um, is there a tool to go from firewall vendor a to firewall vendor b, because there's like a thousand net roles in here and I don't want to hand jam that yeah. And it like sometimes like people take cracks at those tools early on, but then you know both vendors do upgrades and changes and stuff like that and it's seemingly impossible to keep up with a tool that will help you convert from one to the other.

Speaker 2:

Right and that's and.

Speaker 1:

I'm sure like one vendor doesn't want you to convert to another right. So, they're trying to mitigate that aspect as well.

Speaker 2:

Right, and that's where I could see like AI or some kind of machine learning to figure out when an update happens, if any kind of code changes or anything like that. It figures out what that code changes and how to translate from vendor a on this code to vendor b on that code, kind of thing, and in some cases the same vendor.

Speaker 1:

Yes, you know when you do certain code upgrades like it can sometimes mean a world of change.

Speaker 2:

Oh yeah, like uh A sas, that was a good one. Uh, what was it? A, a dot two to a dot three or something like that. I can't remember which version it was, but they like redid net all together on the command line and whatnot. And uh and dan had nine hours of trying to fix bull crap Because? Because I did not know.

Speaker 3:

No, so we've teased this a little bit, but I want to talk about potential privacy implications of leveraging AI. So we talked about how it would be great to leverage these large data lakes of Public information, of course, anonymized. But do you think we'd ever get to the point where we can leverage those, those public data lakes for AI? And one of the things I want to bring up is there was a recent article is chat gpt has been all the rage of employees at a large company that were pasting proprietary code into chat gpt to try to get suggestions or whatever. I'm not 100% sure on the intent, but that's, that's obviously a, a intellectual property in a in a privacy issue. So how do we, where do we, even start with that leveraging AI in the enterprise?

Speaker 2:

So I think Uh, the episode that you were talking about with packet push pushers. Uh, was that greg and uh?

Speaker 2:

yeah, greg, and jonna yeah, exactly, so I listened to that episode and I thought it was really good. They they brought up a really good point about how you ingest your company's data into your own chat, gpt Right and and pull information out of that, and I thought that was, I thought they made really good points with that and uh, and so I feel like that's kind of the only way we can go with this right, like because if you, if you put your your data out there, I mean it's gonna learn that, and then so what's stopping someone from being like okay, let's say, a bunch of nasa Developers put a bunch of their proprietary code out there. Well, what's stopping me from saying what is nasa doing, you know? And then it telling me that, no, exactly.

Speaker 3:

Yeah, that's it's a huge privacy issue and the I do think that having Some sort of private type AI solutions is is gonna have to be the future.

Speaker 3:

Yeah, the only question I have around that is and maybe this isn't a problem for most organizations but the main question I have about that is at that point You're you're leveraging your own private AI solution with your own private data and you only have your own data and maybe for for you know, certain companies, that's fine. All you need is your own data and you need to be able to learn from it to be able to automate certain things. But I I'm still going back to, like security use cases and that you you need data from from other entities, but I guess a lot of that is either public information Already through certain Organizations or you're getting, you're paying subscription fees to get access to that data. So maybe it's it's not as big of a deal, but, yeah, I think you have to have a way to run at least your own private instance so you don't have to worry as much about that loss of intellectual property.

Speaker 1:

Do you think that there's a scenario where, yes, you have a private instance, but it's still connected to the public instance, so it's it's ingesting the questions and the data that you're giving it and then it's using the public to go Reference and look for potential answers, but not putting your intellectual property out into that pool as well?

Speaker 3:

Yeah, one-way street kind of thing, yeah, yeah.

Speaker 2:

Hey wait, are you trying to say that it's going to do NAT of some type?

Speaker 3:

Yeah, I was just kind of. I was trying to equate it to a networking term.

Speaker 2:

Yeah, no, I I think that's a good question, like what kind of data are we talking about, right? Like if if you're uploading your code and if your code has anything that mentions your company name, is there a way for it to To take that out, remove that to where it's getting more about the code and not who the code is for exactly?

Speaker 1:

Oh, I got a new business like sanitizes it before I got a new business venture fellas.

Speaker 3:

We can get in, we can get into, we can be a dlp provider for for ai solutions man.

Speaker 4:

Yeah, I mean that's what it is.

Speaker 3:

It's data loss prevention for for leveraging public data for ai.

Speaker 2:

Wow, the sky is the limit, really. I mean, I, I don't know that. That's why I'm kind of torn on this topic, specifically Because a part of me is like I want to send certain type of data, but obviously you don't want to send out your company's like bread and butter data, right, I don't know it's, it's a, it's a difficult subject in a sense.

Speaker 1:

Now, when I signed up for for chat, gpt and mind you, like I only did it just today, right, because I know myself and I was worried about, like, what's going to happen if I start using this thing, right, like, am I going?

Speaker 1:

to get just like too hooked, and it's not that like I'm not looking for ways to cut corners, it's just like it's. It's very interesting technology, right, and it's like the gaming that we talked about at the beginning of the episode, right, I could totally see myself just getting kind of hooked on on, you know, just playing with this thing and seeing how it works. But what I saw when I signed up for it was, like you know, our analysts will sometimes look at this stuff to help train the AI. So don't put anything personal in there. So it wasn't necessarily from a standpoint of don't put personal information here because it's going to be added to the pool of Knowledge that the AI reaches into. It's just some actual human might see this at some point. Maybe and you don't want that potential human seeing your personal Stuff information, whatever does chat GPT take the information you give it and then remember it and maybe Provide that as a response to somebody else.

Speaker 2:

See, I don't know. I mean, it obviously remembers what you say to it to give back feedback to you, right? The contacts to yeah, right, but that's a good question and I don't know. I Like, can I ask it something that maybe you might have injected data into it and get it in a return on mine? You know what I'm saying.

Speaker 1:

Yeah, right, like if you tell it something personal, can I go to chat to you, pt, and say, hey, tell me that super personal thing about Dan Richards.

Speaker 2:

Yeah, exactly Like this, just took stalking to another level.

Speaker 3:

It's gonna tell you something about a toll booth. I'll tell you that right now.

Speaker 4:

Dan, don't know how to drive.

Speaker 3:

So I see to Just to kind of, as we're recapping what we're talking about, I'm seeing two really big challenges and it's it's the the privacy, potential privacy implications that we've talked about and Really trust in data, because, to me, an AI solution is only as good as the data it has to analyze and make decisions. So you, to me, you really have to trust what data inputs that that AI systems getting, and and I don't have any answers around that on on how you do that, how you sanitize data, how you make sure that You're ingesting what you you need to, and and I don't even want to get into the the next frontier of, you know, potential malicious code and input validation on AI systems, and that's a whole nother can of worms. That is just crazy. But though I mean, to me those are the two big challenges with Leveraging AI solutions and in networking and in just business in general. What do you, what do you guys think?

Speaker 1:

Yeah, I agree. I mean, when I think about AI and the uses that I've seen, like using it as as data lakes to help you troubleshoot something, I think is the the most prevalent use case and Maybe even for the foreseeable Future, right like the the short term, because when I was playing around with it earlier, I was just asking it to build some simple configs using, like, security best practices, and was it right? Yeah, but it was missing some really key information and and it took the parameters that I gave it and like, for example, like I said, build me a switch configuration and on the trunk ports, use native vlan, you know whatever, and you know security best practices and all that, and and it split the switch in half. Now, first of all, I didn't say it was a 48 port switch, but ports like 25 through 48 were all trunk ports.

Speaker 2:

Yeah, but that's where it's nice that you can go back and you can correct that right like sure like you could say hey, by the way, 25 to 48, whatever you said, I don't want those trunked. And it'll be like, okay, give me a minute, and then it spits it back out to Okay, yeah, I mean it now, like, like I said, I just signed up today and, yeah, this was the extent of playing with it.

Speaker 1:

Now I did do some things. Like you know, give me some prompts for writing blog articles and there's some great stuff. You might see some content coming from my blog at some point here. But but I just I don't know if, if chat GPT is the tool that's gonna help me write configs right when as I go to to do deployments, but I I only just started playing with it and I don't know if I want to do that right. Like I've seen some mistakes in Trying to go through and decode and tune and engineer my inputs to always give back the right inputs. I think it's just gonna take too long to find to.

Speaker 1:

I don't know if if that use case is the best use case for something like chat GPT, I think it's a great tool that was another big point in this heavy strategy podcast episode about is chat GPT coming for your job?

Speaker 3:

is that it really seems like chat GPT is. The reason they're making a big push for this is to show people the power behind these types of technologies. Not necessarily that chat GPT is gonna be the end all be all. It's really just the tip of the iceberg in in what is Available on out there and in what we'll be able to see potentially in the future.

Speaker 1:

Yeah, it's exciting and a little bit scary.

Speaker 3:

Oh, I think it's more than a little bit scary.

Speaker 1:

Okay, cuz what I forget. There was like some company that was saying they were seeing the AI do stuff that, no, nobody told it to do.

Speaker 4:

Yeah.

Speaker 1:

I think, they made a movie about that.

Speaker 2:

So I want to pivot a little bit what? What about, like network optimization? What are, what are some ways we think AI could help with that?

Speaker 1:

One thing we I mean, haven't we seen that with Juniper and missed right like hey, you know, look at my network and see the types of problems I'm having and and make some recommendations on how I can fine-tune or make things better? I mean, I think that you see a lot of that with like wireless and DNA center and I think we saw some good use cases for that with with Juniper.

Speaker 3:

Missed product Gotcha yeah, I think one thing, makes and when that's just what I was gonna say, dan. I think one thing that there could be a major use case for for AI and I know there's solutions that are out there that are doing similar things already is potentially leveraging AI as a Capacity planning tool to be able to help tell me. Hey, similar to the SFP use case that we brought up earlier. Based on the data that we're seeing from this SFP off of this port in the switch, we think it's due for failure in X amount of time.

Speaker 1:

Similar to that.

Speaker 3:

Yeah, to be able to capacity plan. Hey, we're seeing it's getting ingesting all this data. We were seeing your network grow. In these parts of the network we think you'll be at 95% utilization of this link in X amount of months you may want to do something about it, like, hey, you need to start a project of figuring out how to fix this exactly.

Speaker 2:

Yeah in exactly three months.

Speaker 3:

Yeah, oh.

Speaker 4:

By the way.

Speaker 3:

Here's some.

Speaker 2:

Here's some switches you may want to buy Right no and like do you know? My favorite thing is is when something does break. You know, our management wants us to do a kind of like a root cause analysis right, and I absolutely love those. You know. It's just the. The joy of my working career is doing a root cause analysis on why something messed up. I could see where AI could help help it yeah, right yeah.

Speaker 2:

Yeah, for your point of you know, if that SFP failed, it could, it could say look, this is why it failed. You know, the thing is 12 years old. Move on, guys.

Speaker 3:

Yeah.

Speaker 2:

You need to. Just it's a $200,.

Speaker 3:

You know piece of gear, get a new one and I think it all goes back to and I harp on this all the time that it departments are Are seen often as cost centers and that they're really just there keeping the lights on. And I think AI technologies can help be that differentiator, to be able to show that you can provide value, because you're not just Operating the network, you're you're not just letting it run, you're letting it provide suggestions, you're letting it try to make things better on its own and suggest how you, as an engineer, can get involved to Make things better in the long run and enhance the network over time. That's where I see AI, as far as building networks and not just the, the operational hey, you may have problems here and there, but actually building and enhancing. That's where I could see AI being helpful in the future is doing that capacity planning and saying you know, hey, giving ideas for optimization, and where you can enhance the network.

Speaker 2:

Yeah, and then you know we talked about this at the juniper thing. But like, what about self healing? Right, yeah?

Speaker 3:

being able to make Automated changes. And that's back to that trust. Right, you got to make sure that you trust your AI system to accurately define something as a problem one and then accurately implement the solution Without a human being consulted. Now, personally, I think you know, for general adoption, I think we're a little ways away from that, but I don't, I don't see that not happening down the road I.

Speaker 2:

I want to get to a point where, if let's say that SFP was gonna fail, right, let's we kind of keep using that as a as an example here. Yeah, I want AI to be able to go on Amazon, order a new SFP, deliver it to my door, you know, and then I can go install it.

Speaker 3:

I Thought you were gonna take it a step further and you want the switches to have like an eject button a little.

Speaker 4:

I Think that's, that's a lot closer than you think.

Speaker 1:

Dan, I mean based on conversations we saw, particularly NFD earlier this year. You know to proactively let you know like, hey, based on on the performance that we're seeing, this has like, the signature of this is going to fail. We're gonna send you a new one, you know, like that's, I think, within the next few years, that's, that's definitely gonna be part of, like you know, your next service contract and that's what you want, right?

Speaker 3:

You don't, you don't want your, your customers, to have to open up tickets. I mean, that's, that's a win-win for For these vendors, right? Because then if there's automation solutions that are Doing that, for instance leveraging AI, then not only is that a value to the customer and that they didn't have to see that that we have a problem and report to the vendor that we have a problem, but that's also taking less time from the tack team at the vendor to take the call from the person or respond to the online ticket Process, the rma. Well, first troubleshoot and make sure that it really is, because you always get that right. Hey, I've got a bad power supply, I got a bad SFP, whatever. You typically got to go through the steps with the person. Hey, did you check this? Did you check that? You check the third thing? It could potentially, to your point, aj, take all of that away.

Speaker 3:

Oh yeah and yeah you come into work and there's your SFP sitting on your desk and you just need to go swap it.

Speaker 1:

Yeah, I've seen like this proactive stuff before. What one thing that comes to mind is like Data storage right, like hundreds of drives You're gonna have failures, and they would always say proactive right Like the drive failed, we open a ticket, we're sending you the rma, right Like. But now we're talking about Notifying me before a potential failure and I think that that's like a whole, that like that's actually Proactive, like there's like automatically reactive versus proactive.

Speaker 3:

Yeah, there's not a problem now, but we're seeing indicators where there could be and that's where I think we can leverage Dan, you brought it up earlier with the Windows example and you can opt in to send an optimized data of your what Windows sees as being your, your user experience. If we have something like that in network operations to where you can tap into Unknowingly that the system is just doing it for you, you can tap into those data lakes of other customers that are experiencing those similar problems. That it's not just your own data. You, the vendors, are able to ingest all of that and make decisions based off of that. Now, all of a sudden, you have indicators of potential issues from all the different, or most of the different customers that that vendor has. That makes life better, better for everybody in the end, as long as it's, you know, done safely and securely, of course.

Speaker 2:

Yeah, and that's where I feel like if there's any company out there that has any form of a security team, like that ain't gonna happen. I just don't see him wanting to allow that. Yeah you know me personally, it's like whatever. If it's secure, then why not feed that data out there to help everyone? Essentially?

Speaker 3:

And AJ's right, we're already there in some cases, or very close. I mean we were seeing stories about that at NFD and talking about it at that Juniper event too. I mean this is very top of mind for a lot of these vendors and it's nice to see that they're looking that way, they're working toward that way and things are actively starting to get better, because anytime that an issue can either get resolved before it actually becomes a customer noticeable issue or a ticket can get auto-generated with all the information because I do that, I try to train myself but I do this shit all the time where I know I put in a case with a vendor and I know that I need to put the certain three outputs or four outputs. I know that they're gonna want but I forget Show tech and that's the very first thing they ask. So to be able to have a vendor have that, that level of information already and be able to correlate, it is huge.

Speaker 1:

Well, guys, this has been a fun episode. Any final thoughts before we wrap this one up.

Speaker 3:

We should have just had chatGPT. Do it for us.

Speaker 1:

Who do you think is gonna?

Speaker 3:

race for us, maybe it did.

Speaker 2:

Yeah, who's to say it didn't? Also, I wanted to throw this one joke in, but back when we were talking about more of the ethical you know that kind of stuff and security and whatnot, you know what if the network sees a user, let's just say his name's Jim. Okay, jim is putting in tickets saying that, hey, the network sucks. You know, I'm having a horrible experience here. What if AI sees that and it takes offense to it and it's like you know what, jim, you're gonna get the absolute worst experience now and it starts picking on users like you've never had a bad experience before you.

Speaker 1:

Just weird yeah.

Speaker 3:

It turns into the grumpy dad wearing the wearing the Jorts and the in the Moan shoes.

Speaker 2:

Every time that Jim tries to open a ticket now, for whatever reason, it just goes, sends him to YouTube, to Rick Rold, you know Nice.

Speaker 1:

Well, this has been a fun conversation. I would love to hear from you, our audience, on what you think the future of AI and networking is. Let us know in the comments if you're watching this on YouTube or if you're listening to it on the podcast. Jump on Twitter, mastodon or whatever social networks hot these days, and let us know. We'd love to continue this conversation. If you haven't yet, please check out our latest podcast, cables to Clouds. You can join Tim, chris and Alex. Alex.

Speaker 2:

Say it again, say it again, we'll cut it out, sorry.

Speaker 4:

Alex.

Speaker 1:

You can join Tim, Chris and Alex every other week, every off week, from the A1 show on their new podcast Cables to Clouds, where they talk about all things cloud networking, and it's a fun show. Another fun show and, as always, if you're enjoying what you're listening to, please share this with another network engineer and help the show get more ears. Thank you so much for joining us and we'll see you next time on another episode of the Archive Network.

Speaker 4:

Hey, dear friends, we hope you enjoyed listening to that episode just as much as we did recording it. If you want to hear more, make sure you subscribe to the show and your favorite podcatcher, you can also give that little bell rascal a little ring, a dingy, so you know when we release new episodes. If you're social like we are, you can follow us on Twitter and Instagram. We are at Art of NetEng, that's Art of N-E-T-E-N-G. You can also find us on that weaving web that is the internet, at ArtOfNetworkEngineeringcom. There you'll find our show notes and some blog articles from the hosts, guests and other friends who just like getting their thoughts down on that virtual paper. Until next time, friends, thanks for listening.

People on this episode

Podcasts we love

Check out these other fine podcasts recommended by us, not an algorithm.

Cables2Clouds Artwork

Cables2Clouds

The Art of Network Engineering