Not too long ago, it seemed everyone's plan was to lift and shift an entire data center's worth of workloads to the public cloud. Today, the industry is starting to rethink that strategy and be much more thoughtful as to what should go in the public cloud, why, and how. In this episode, Ted Turner, a cloud solutions architect at Kentik, joins Telemetry Now to talk about the changing narrative around moving everything to the public cloud.
Digging into applications and infrastructure to get the best performance for customers. Specialties: Co-ordinating teams to get the best results out of networks and server architectures. Hands on building network and server architecture.Connect with Ted on LinkedIn
Phillip Gervasi: This is Telemetry Now and I'm your host Phil Gervasi. And with me today is Ted Turner, a cloud solutions architect with Kentik and a veritable cornucopia of knowledge when it comes to public cloud and the journey and struggles that many of us have gone through in recent years trying to lift and shift from our own data centers into the public cloud. Now we'll be answering the question, is the cloud first initiative that's been coming from many CIOs and CTOs lately, always the best path forward? We're going to be talking about the real reality of just picking everything up and tossing it into AWS, Azure, or GCP. So let's get started. Hey Ted, great to have you on today. I'm really looking forward to speaking to you. Before we get started, would you mind introducing yourself to the audience, a little bit of background about who you are?
Ted Turner: I'm Ted Turner, cloud solutions architect over at Kentik. I've been in the business helping manage applications and networks for close to 20, 30 years now.
Phillip Gervasi: 20 or 30 years in the business. All right. Well you certainly are a veteran then. One of the things that I do want to say that I appreciate about you, Ted, having gotten to know you is we sort of have a very similar background in that we both started working with SMB at first, small and medium business and then got into enterprise networking and then bigger and bigger kind of global scale enterprise networks at some point. So that is a cool commonality that I appreciate about you. Now I got back from a short camping trip in upstate New York recently and I was very proud of that. So I'm kind of bragging about it on the work Slack. And then Ted chimes in with a recent hiking trip that he did. So for our audience, where did you go, Ted?
Ted Turner: I was in South Korea and Seoul, Korea and I was up on top of Namsan. So it's a south mountain right there in the center of Seoul, Korea.
Phillip Gervasi: Wow, that really sounds amazing, Ted. I mean I do like taking the canoe up on some little pond in upstate New York, but I will concede that you win this round. Okay. So I want to jump right in here and talk about kind of the elephant in the room about public cloud. So I remember years and years ago, not that many years, but a few years ago, the big thing when leadership started talking about moving their data centers into public cloud was cost savings. Right. It was this idea that we are not going to own anything anymore. We are going to pick everything up, put it into AWS or Azure, whatever, and save tons of money and then all get huge end of year bonuses. And that's not exactly the case anymore though, is it?
Ted Turner: No. So I saw a case where we took our FAQs and we had call centers taking phone calls, answering questions all the time, and we loaded those out into AWS and all of a sudden the call center telephone traffic dried up. So I have seen huge cost savings materialize, but then you start to move the rest of your applications, not static content, and then that's where the rest of the conversation starts.
Phillip Gervasi: And by the rest of the conversation you mean the conversation around which applications make sense to migrate into the cloud because they'll perform better or because they'll be more accessible and ultimately cheaper to run in the cloud. But not always.
Ted Turner: That's not true for probably 95 to maybe 99% of your business.
Phillip Gervasi: What? I mean 95 to 99% of my business, that flies completely in the face of pretty much everything that we've heard about migrating all of my workloads to the cloud and saving a ton of money. I mean that was the narrative and the main impetus that people started that process some years ago, right?
Ted Turner: So for the first three years, our mantra was, we're going to move to the cloud, we're going to start to be faster, we're going to iterate, we're going to be agile, and we're going to save money. After the third year management from the CEO on down said cost savings, take it off the board. That is not something we're going to measure.
Phillip Gervasi: Yeah. I remember having similar conversations about SD- WAN not long ago. When SD-WAN was brand new, 2016, 2017, one of the main drivers was cost savings, get rid of your NPLS links and some other kind of, running over the internet and getting rid of lease lines and private lines. But soon after implementation started rolling out, the reality set in, is we're not going to really save much money because a lot of the time we're keeping our NPLS or we're locked into our NPLS because we're under contractor or whatever other reason, it's just, it's the same thing. Cost was taken off the board. But with regard to cloud, is it that we've been duped into thinking that going to the cloud was cheaper or is it that the implementations are now more mature and so we're seeing the reality of how cloud is being utilized today, which may be different than the way we thought it was going to be used at first?
Ted Turner: So the first part is when you go into the cloud, you can build it right. If it doesn't work, you can start from scratch and build it over again. That's the cool thing because most of the time you still have your data center. So where we came from, we were pushing boxes of software out the door back in the day. And so Walmart, Office Depot, Office Max, that software went out annually and we didn't care because we didn't deliver software through the internet. Then 2010, 2011 came, the internet became a big thing and our backups would take three days to do a restore. And so having that data replicate between Plano and California was impossible to do with those same circuits that you were talking about, those NPLS circuits. It was just impossible. When we moved into the cloud, we could do those things. So now when we're doing this high capacity, making sure that the customer data is backed up and in multiple regions, that starts to become a problem. And that adds to your complexity, that adds to your cost. So just from that basic premise things that as your company grew, you didn't necessarily do in the data center before. Now you can do these things in the cloud, there's nothing fettering you and you can actually do your well architected application, your well architected frameworks, you can build it correctly. But when you do that, you're incurring costs, costs that you avoided maybe in your data center before.
Phillip Gervasi: All right. And so then it sounds like the nature of how industry is utilizing cloud technology has changed. It's not just lifting and shifting a VM into the cloud and then running it from there, but it's all the complexity of how those resources are consumed today, distributed applications, multi- cloud containers, and also the inherent cost, high cost of getting to and from the cloud over the public internet. But wouldn't you say that maybe some of that cost is worth it considering the incredible benefit that we have to availability, resiliency and that sort of thing?
Ted Turner: It is, but it's costing more. But if you went and built that in your data center, you're going to drive your data center costs up sky high. So now you can start to take a slice of your business at a time, but you're not actually saving money, you're spending more.
Phillip Gervasi: Right. So I think I understand your line of reasoning. I'm going to wander a little bit here in my explanation, so just bear with me. But ultimately what you're saying is that applications and service delivery are much more complex than they were years ago. They're more complex today and there is an inherent cost to that, whether you're deploying and running stuff from on premises in your own data center or in the cloud. And so it's by virtue of the complexity and the way we consume applications that we have a high cost, not necessarily because cloud is just crazy expensive and we were duped. Now, I guess that begs the question, if our strategy needs to change because the way we use applications necessitates that, assuming that we don't have an unlimited budget, how do we go about determining what we migrate to the cloud? Which workloads, which applications and how do we determine that from a cost perspective as well?
Ted Turner: You're going to have to take several attempts at this. This is kind of the key learning as you're trying to put things into the cloud, trying to take pieces of your application stack, whatever it is in your business, take something simple, maybe identity management, maybe just document storage, something simple and start putting out into the cloud, making sure it's secure, ready to go and see where those costs start to pile up. There's a lot of ways to go out there and save money in the cloud, but those things that you're used to for your business, making it available for your staff, making it available for your customers, they may have to change some of what they do. So it may have been you're using a computer and you're using Windows filing system and now you have to use a web browser to gain access to some files. There's a change that needs to occur there. And so that inured some costs because you've got some training, you've got people who are resistant, you have people that want to do things the old way. Why do they want to do these newfangled methods? So there's just a lot of complexity in enabling that change, let alone when that data starts to hit. We just did a cloud field day and one of the customers had data writing from their data center out to a storage location in the cloud. And when they moved that same application from the cloud to the cloud, they're moving across the public internet but still in the cloud. And they didn't realize that transferring data out of the cloud and then back into the cloud is expensive and it was going to cost close to a $100, 000 a year simply because that migration from on- prem to the cloud, they're thinking the same way, but it doesn't actually do the billing the same way.
Phillip Gervasi: So in that particular case though, was it kind of the best choice? Cost aside, was it the best technical manner to move that data around and therefore worth the 100 grand a year?
Ted Turner: So in the cloud you just don't move it across the public internet. You stay within the cloud provider and they drop their cost dramatically, close to 90%.
Phillip Gervasi: Okay. So there is a strategic way to move applications into the cloud and maybe drop that 99% of applications down to 95, 92. But you have to be very strategic about it so that way you're identifying where you're incurring your costs. And correct me if I'm wrong, because so far it sounds like a lot of the costs that you're going to incur aren't necessarily like I spun up these VMs or these containers and I have a cost now, a monthly cost. It sounds like a lot of it is from the movement of traffic, of packets and flows, right?
Ted Turner: You got it. It's storage, storage of that data so you can pull it up this week, next week, two months from now as well as the transfer of that data.
Phillip Gervasi: So it sounds like the nature of the visibility that we need into our infrastructure and into the traffic that's going to and from our resources has changed. I mean there was a time when all I really cared about as far as visibility was an up down status of my servers. Whether they were bare metal or VMs, was it pingable, accessible? And that was good enough. Maybe I cared about latency a little bit and if there was voice involved jitter, things like that. But ultimately it was not much more than that. Whereas today, yes, we're concerned with that, but we're also very, very much concerned about how much traffic is going to specific cloud resources in the context of cost, including how I get to those resources, right, the ingress and egress to AWS and also the path that I take to get there when you consider the cost of transit networks.
Ted Turner: You got it. There's the backend links into the backend. So your express routes and your direct connects, that can lower some of your costs. Anytime something's going over the public internet, there's that premium of going into and out of your cloud provider.
Phillip Gervasi: Okay, so then how do we do that? Is it just collection of flows, flow data?
Ted Turner: It is collection of flow data, NSG flow logs in Azure VPC flow logs in AWS, same, similar in GCP. And you can see volumetrically what things look like. What that pathway is, is going on the backside, going across a fiber path into your data center, into your campus, or is it going across the front side. And things that are going across that front side, across that public internet are going to be your biggest, it's going to push your costs up.
Phillip Gervasi: So we've wrapped this entire conversation of whether moving to the cloud for specific applications or all my applications, we've been talking about this in the context of cost more than anything else. Is it cost effective today? And really we haven't looked at performance. So I get that we can see how much traffic is going to what resource, we can look at flows and we get a lot of network telemetry that we can mine out of that. But what can we do as far as performance? I mean is it wise to move an application to the cloud in the context of performance? Will we have the same end user experience or will we have a degraded experience and therefore maybe moving that application to the cloud's not the best choice.
Ted Turner: You got it. Those flow logs will tell us how much data was going into and out of that server, but that's not going to tell you what performance looks like. So what that cost is going to look like, flow logs will help you. But performance wise, what that looks like for that end user, that staff member trying to access that web server, that's going to be synthetic testing. So test geographically from wherever your staff, your customers are coming from and then testing against that web server that you were just talking about.
Phillip Gervasi: So it sounds like in order to have a modern effective strategy for migrating applications and services to the cloud or from cloud to cloud or from the cloud to back on premises, we really need the visibility into, from a volumetric standpoint, so we can see how much traffic is going where. And that's a big contributor to cost. Right. We also need the performance monitoring so we can know how effective a perform or rather an application is performing from an end user's perspective in a particular cloud or in a particular or specific region. Right. So then we can determine whether it's better off in a different cloud or back on premises. And in that sense, we really need a holistic approach to visibility to see both from a cost perspective, from a performance perspective if it is viable and really the best choice to migrate applications and workloads to the cloud as opposed to just take it all and stick it into AWS.
Ted Turner: You got it. That's why we're starting to call this stuff network observability. It's tying all the different pieces together from all the different areas. So those DNS requests, how long is that taking? That's gathering that webpage, that HTML page that's coming down and all of the graphics that come with that. How long did that take? What did the TCP connection take? Tying all those things together with the flows helps you start to understand holistically, what does that customer experience look like?
Phillip Gervasi: Yeah, I can't believe I didn't think of mentioning DNS. I'm almost ashamed considering how many times that's bitten me in the rear end. You know what I mean? And now more than ever, because I have like 77 different DNS servers redirecting to each other and half of them are in the cloud. And that's why the application's slow, right, because of some random DNS server somewhere that's taken forever or is pointing to something it shouldn't be, so.
Ted Turner: We had a deployment in Australia and they were using DNS in San Francisco and they couldn't understand why geolocation wise, all of the files were being downloaded from the San Francisco Bay Area for Australia and just driving everybody batty in Australia.
Phillip Gervasi: Wow. Yeah. And it was because of misconfigured DNS, something not working properly.
Ted Turner: Misconfigured DNS.
Phillip Gervasi: Yeah. Yeah. And as much as we like to joke about DNS being the root of all our network problems, I mean it really is an important piece of our infrastructure. I mean that is how many organizations, if not most load balance across the global internet to identify where you're coming from and therefore which region you should be to redirect it to you to get to whatever resource you want. Right. That makes sense. And which I have to assume is tied to cost as well, right?
Ted Turner: You got it. Your content delivery networks, that's exactly how they work. And Amazon, they call it Route 53, that's just their DNS, but they're doing all of the routing for you through DNS routing.
Phillip Gervasi: Yeah, it's pretty cool how that works on a global scale. I've looked into that before for troubleshooting and for design purposes, but I'm just reflecting on how technology changes over time. It's so interesting to me because I remember only a few years ago when everybody had a cloud first initiative and it was really just take everything that I have and put it into public cloud and own nothing and then just pay for everything as a service. And how that's just changing as a result of cost and performance issues and how we're just becoming more strategic in what we move when and how. It's kind of like with SD- WAN, I'm bringing that up again, how the initial impetus was to save money on ditching NPLS and then that narrative changed as well over time as everything matured and we looked at the reality of how things were implemented and then utilized and consumed after the fact. So anyway. So Ted, I think right here is a good place to stop. And before we close though, how can folks reach you online if they had a question or wanted to comment on something from today's show?
Ted Turner: I'm on Twitter. I'm Ted Turner in Cal.
Phillip Gervasi: Okay, great. And you can find me on Twitter at network underscore Phil. You can search my name on LinkedIn, Philip Gervasi and pretty active in both places, so. And to listen to other episodes of Telemetry Now you visit our website, kentik. com/ telemetry- now. And if you're interested in being a guest on Telemetry Now or you have an idea for a show, you can email us at telemetrynow @ kentik. io. We'd love to hear from you. So until next time, thanks for listening. Bye- bye.
Do you dread forgetting to use the “add” command on a trunk port? Do you grit your teeth when the coffee maker isn't working, and everyone says, “It’s the network’s fault?” Do you like to blame DNS for everything because you know deep down, in the bottom of your heart, it probably is DNS?
Well, you're in the right place! Telemetry Now is the podcast for you!
Tune in and let the packets wash over you as host Phil Gervasi and his expert guests talk networking, network engineering and related careers, emerging technologies, and more.