Follow me:
Listen on:

Day Two Cloud 102: Edge Cloud Isn’t Magic

Episode 102

Play episode

Today we’re talking Edge Cloud. What does edge cloud mean? Is it the mashup of two buzzwords to create an even more powerful buzzword? Is it something more than just a little data center?

Our guest Alex Marcham has written a book on the subject and we’ll get into his take on edge infrastructure, what edge cloud is all about, real-world use cases, and how it differs from typical colo facilities or centralized public cloud data centers. We also look at essential concerns for edge deployments including networking requirements, 5G, and the workloads driving edge infrastructure.

Alex Marcham is Director of Edge Service Innovation at Digital Realty and the author of Understanding Infrastructure Edge Computing, published by Wiley in April 2021.


  1. There are many different tiers of edge data center and infrastructure
  2. Edge cloud is many things to many people; be specific in what you need
  3. Interconnection needs to be understood to make the most of edge cloud

Sponsor: CBT Nuggets

CBT Nuggets is IT training for IT professionals and anyone looking to build IT skills. If you want to make fully operational your networking, security, cloud, automation, or DevOps battle station visit

Show Links:

Understanding Infrastructure Edge Computing – Wiley

Alex Marcham on LinkedIn



[00:00:00.240] – Ethan
[AD] Sponsor CBT Nuggets is IT training for IT professionals and anyone looking to build IT skills, if you want to make fully operational your networking cloud security automation or DevOps Battle Station, visit CBT nuggets dotcom cloud. That’s CBT nuggets. Dotcom cloud. [/AD] [00:00:25.780] – Ethan
Welcome to Day Two Cloud. And today we are talking Edge Cloud. Edge, is it a word, what does that mean? Is it hype? Is it buzz? Is it something different than just a little data center or what?

[00:00:36.520] – Ethan
Well, our guest, Alex Marcham, has written a book on this very topic. We’re going to get into his take on what edge infrastructure is all about, what edge cloud specifically is like, whether or not you should actually build such a thing and what the use cases are. And Alex is a fountain of knowledge. Wouldn’t you agree Ned?

[00:00:52.480] – Ned
I certainly would. I threw a lot of questions that may not have been in the script at him and he performed admirably. He has a a real breadth of knowledge across the entire topic, around Edge Cloud and and things that are adjacent to it, like 5G networking. We get into that a little bit and how it impacts networking with Edge cloud. So I think it was it was a very interesting conversation. And I just, man my brain wants to fall out of my ears right now because so much stuff.

[00:01:21.880] – Ethan
I know exactly why you say that. And in about two minutes you will to strap in for this conversation with a book author, Alex Marcham, Understanding Infrastructure Edge Computing is his latest publication.

[00:01:34.210] – Ethan
Alex Marcham, welcome to Day Two Cloud. And you’re here because you wrote a book, man, this is this is all on you. If you hadn’t written this book, I don’t know that you’d have to be doing this interview. But but here you are. You wrote a book. Tell us what this book title is and what it covers.

[00:01:49.190] – Alex
Right. But first, I got to do my author joke, what do you call a pedant that write things down, an author. That’s my favorite author joke. The only one I’ve got. That’s what I’m going to say. So the book is called Understanding Infrastructure Edge Computing wrote it with the lovely people over at Wiley. And essentially this is focused on a specific tier of edge data center infrastructure. So this is focused on what I’ve been referring to over the past few years as the infrastructure edge, which is a tier of physical edge data center facilities, anywhere from sort of 500 watts to 250 kilowatts.

[00:02:22.180] – Alex
I think of the typical shipping container kind of form factor edge data center that you’ve seen over the years. So concerned with that. What are the constraints, what are the benefits of such a form factor, anywhere from vehicle mounted through to, as we’ve talked about a moment ago, shipping container size. What are the network characteristics that make a particular site or location useful and everything else in between? So it was very fun to write and I hope enjoyable to read.

[00:02:48.970] – Ethan
So if I’m putting a data center in that container form factor, that’s the context of what you’re writing about through this thing.

[00:02:56.230] – Alex
Yeah. And the various factors that you have to understand, you know, a big one in there for me is don’t just put a data center in there and drop it in the middle of a field in Kansas because the network connectivity, both on the front haul and the backhaul part, are really what makes those different deployments actually valuable. So that’s that’s a big, big point of contention in the book.

[00:03:16.640] – Ethan
Okay. In the industry, we have a lot of fun with the term edge. What does that really mean? Edge Cloud has been coming up a lot. Does this come under the heading of Edge Cloud?

[00:03:29.050] – Alex
In a way, yes. So to my my personal definition of an edge cloud is a set of cloud services that are deployed in physical, geographically distributed data center locations, of which some of those could be these infrastructure data set of facilities that the book talks about, if that makes sense.

[00:03:50.310] – Ethan
Yeah, if I’m if I’m right, I’ve got a form factor. That’s this container, let’s just call it. And if I’m serving up my infrastructure via cloudy methods, then yeah, sure. We’ve got we’ve got edge cloud here.

[00:04:04.780] – Alex
Yeah. It’s, it’s one of those very. Very imprecise terms, right? I mean, edge itself, as you, as you just mentioned, is very wishy washy in its own way. Cloud still is in a lot of ways too, so you put the two together and you’ve got kind of a recipe for disaster, a recipe for imprecise communication anyway. So, yeah, I would put it anywhere from a set of data centers that are distributed perhaps on on a like a subregional level compared to what we look at from the major clouds today, I’m not talking US east, US West, maybe Chicago or St. Louis or Salt Lake City, that kind of level down to potentially because I live down here in Arizona, could be like Phoenix north, south, east, west, depending on how small exactly those data center facilities are going. But that’s kind of my general cutoff point from what we think of today is like centralized cloud versus edge cloud. It all gets very confusing.

[00:05:01.890] – Ned
It does. What I think of centralized cloud, I think of the big public cloud providers. So AWS, Microsoft Azure, Google, they have these big regional data centers. Sometimes, I don’t know, it’s, you know, two or three data centers in each region, but still they’re relatively close together and they’re not situated to provide edge services. They’re just there because it’s a convenient location that’s close to a lot of networking, like you said.

[00:05:27.360] – Ned
So I guess the differentiator here for me is the size and locations of the data center. When you’re talking about cloud, are you speaking of cloud in the operational context, like you’re performing cloud type operations as opposed to the way people usually think about the cloud, which is just some big data center that I don’t own?

[00:05:46.290] – Alex
Yeah, I’m thinking it more in the former, as you said, like the kind of cloud services that would be provided on those distributed physical data centers. So I’m thinking what you said kind of brought to mind something interesting is do you think of AWS as an edge cloud if they have a deployment in an edge data center down the street from you? Or do you think of an edge cloud as a specific kind of tailored only edge facility level cloud? If you can see what I mean.

[00:06:15.540] – Ethan
Alex, well I mean, it’s a latency problem, isn’t it? Couldn’t couldn’t we say if we want to come up with a metric, whether something is edge or not? It’s more about how close it is to me. As opposed to oh, boy, it does get fuzzy, doesn’t it? But you see the point of making if we use latency as a metric and it’s close, then is that edge?

[00:06:34.060] – Alex
Yeah, that’s that’s a lot of a lot of what I talk about in the in the book and what I was kind of getting at with the network connectivity piece there, because obviously from a from an access network standpoint, let’s assume I’m in office building or residential building, I’m connecting to the Internet or the data center or the cloud, whatever I want to call it, via some type of access network. Right. Cable, cellular, whatever. At some point that has to go to some location to provide interconnection.

[00:07:00.790] – Alex
And then once it goes through that, I can get my traffic off into that cloud service of my choice. So if I have my edge data center, it’s a mile from my house, but my access network runs it 100 miles west and then it trombones back. Right. It gets very complicated because physically we could look at it and say, well, that’s close to me. So it’s Edge Data Center. But by your latency definition, it would actually be worse than the current regional central data center.

[00:07:26.110] – Alex
Right. So unless you have those networks running through that Edge data center, that infrastructure, its data center, whatever scale it is, then it’s kind of kind of missed the boat. Right? If you go by your latency definition.

[00:07:37.060] – Ethan
Maybe then trying to define edge so strictly is foolish and what it comes back down to, as always, it depends. And in this case, it depends on use case. So. So let’s. Well, oh, Ned. I see you got your fingers up.

[00:07:51.280] – Ned
I think I want to go back to the latency thing that you’re talking about, because I do think that’s a useful metric. But you were think of it in terms of just you. And I think if we’re going to define edge in terms of latency, then you need to measure the average latency to any endpoint within a given geographic region. So if I’m looking at Chicago and I’m testing the average latency from 16 different points in Chicago, as long as it’s under.

[00:08:17.940] – Ned
One millisecond maybe, or something like that to reach one of my edge locations, then, OK, I think that really does qualify as edge. It’s that average latency. It’s not just the one specific site compared to where I’m standing right now.

[00:08:32.430] – Alex
Right. But even even in that case, you know, to see any improvement from your edge sites over the existing big, big, bad downtown data center site that’s already there. You’ve still got to have those access networks doing local breakout or other interconnection in those facilities. Right, right. Right. Otherwise, you’re always going to be tromboning through the interconnection point, right?

[00:08:55.240] – Ned
Yeah, if my device is a cell phone, let’s say, and that means I’m going to hit a cell tower, which is going to go to whatever exchange that cell carrier is using, which, like you said, might be one hundred miles away. And then I might have to come back down to that big bad data center and then out to the edge location. OK, that doesn’t make any sense.

[00:09:13.960] – Alex
Yeah, yeah.

[00:09:14.730] – Ned
Which we need to do is local breakout of some kind.

[00:09:17.390] – Alex
Exactly. And I bring that up just because it’s what I talk about a lot in this book is that it’s a lot of what I’ve seen in the market where people get a container data center, put it in a car park and they say, right, that’s that’s that done. And, you know, not really because again, you know, to Ethan if you go by your latency definition, you’re actually performance negative versus the existing facility.

[00:09:39.960] – Ethan

[00:09:40.590] – Alex
You can’t make the same sort of scale savings that you can in a larger facility in terms of power, space, you know, that sort of thing.

[00:09:46.770] – Alex
So you’ve got to assume that you will be more expensive, typically on a cost per kilowatt basis. So if you don’t have the performance improvement, then it’s kind of difficult to justify using it in many cases. Right. Unless it’s there’s some cases where that’s not true or it was kind of an extension of an on prem data center. Like maybe you have an on prem facility, like in a manufacturing center and you just need some 20 kilowatts of supplementary capacity.

[00:10:09.810] – Alex
And it’s only for your local use from that facility. You can run some fiber and conduit across the car park. That’s kind of one use case for those. But for the for the cloud discussion here, that doesn’t really hold up. You need you need multi tenancy to be a real public cloud, I would say.

[00:10:24.220] – Ethan
So one of the edge clouds. Speaking of multitenancy, that I’ve seen one of those architectures I’ve seen put out there is really aimed at service providers. Hey, service providers, you’re going to run an edge cloud. You’re going to take this compute stack. And using the vast resources of your huge network, you can put these stacks wherever you need them. Your tenants are going to come up because they’ve got compute. They want to run low latency close to some set of users.

[00:10:50.250] – Ethan
Augmented reality comes up as one of those fantastic future use cases, gaming maybe being a little more realistic. Would that, in your mind, be edge cloud?

[00:11:01.370] – Alex
I think so, yeah, because you’ve got potentially both both elements, right, you’ve got a set of cloud services that can be deployed from geographically distributed physical data center locations, and you have the ability through the service provider network to deploy those locations across the network, ideally in places where they’re able to do that access network breakout that we talked about. So you could do both of those both those elements. So I would say so. I think it gets interesting when you talk about who’s going to be providing it, because I do think and I know we’ll talk about this a little bit later on, I’m sure, but it kind of gets to the point of operational consistency for me.

[00:11:40.370] – Alex
It’s like in many places, you know, you’ve still already got the in-house or the local stuff that’s kind of versus the cloud presence that the enterprise has. And then do you need another one? Like do you need in-house versus cloud versus edge cloud, all running different systems with IT desperately trying to keep them together?

[00:11:58.120] – Ethan
Operational consistency, OK. Yeah. You say consistency in my brain started fixating on that as like a distributed computing problem. But you mean literally operating this these different stacks and being able to keep ourselves sane as IT practitioners?

[00:12:12.420] – Alex
Yeah. Yeah. And, you know, how do you justify the spend on one of those versus the other. Right. Then it gets kind of gets interesting from an IT managerial proposal standpoint then. Right. Because already having the do I do this in-house versus cloud discussion, if you add a third element to that, which is what do I do in edge cloud, I think it gets much more difficult to have those conversations if the edge cloud is not treated as an extension of your existing cloud, which is why I bring this up with your question of who should the should offer the edge cloud.

[00:12:41.790] – Alex
Like if I get my cloud services already from AWS and then let’s say T-Mobile, just to make up a name, is going to offer me some kind of edge cloud, is that going to be common enough to really be useful to me and provide that kind of consistency that I’m going to need? Or is it going to just add additional headaches for me?

[00:13:00.620] – Ethan
You just made something click for me, Alex. You said extension of my existing cloud. That is, don’t think of an edge cloud as an island that operates in this, you know, place that’s, you know, latency close or however we want to define it, but it’s got to be part of what I’m already doing, just picking up the same compute we’re familiar with and moving it over, which I don’t know, so does Outpost’s and whatever Azure stack, whatever the local thing is there that they call it is that edge cloud.

[00:13:33.350] – Alex
Well, now you’re asking. Yeah, I think it’s I think it gets really interesting when you start going to this specific implementations because, you know, there’s many different ways, even within within AWS that you could do it. I think probably I would put local zones in the edge cloud bucket before I’d put Outposts. Personally, I think Ned’s Ned’s nodding.

[00:13:55.760] – Ned
Yeah, because the local zones are still managed by AWS, whereas the outposts are kind of managed by you. So you provide the space or you’re going to work with like a colo to provide the space for AWS outposts. And then you kind of manage the hardware as opposed to local zones where it’s just, add another zone to my AWS deployment. And suddenly I’ve got four locations all in L.A. and that’s that could be useful. But of course, they don’t have that across the country yet.

[00:14:26.030] – Alex
Right, right. Yeah, it’s I mean, outpost and local zone gets really interesting not to get too deep on the AWS part itself. But, you know, I could I could say from the standpoint of a colocation provider, I have a colocation facility, I run a data center. I could have AWS deploy a local zone, which they manage in that facility, in that same facility I could have end customers deploy outposts in that facility that they manage.

[00:14:54.710] – Alex
So it gets it gets a little bit complicated. And then you could have an on ramp to AWS core locations in that same facility. So it’s kind of to me, it’s that that’s a big part of making this operationally useful and a big part of what I kind of go back to in in the book. And a lot of the time when I talk about edge in general is it’s not as special as you think it is.

[00:15:15.580] – Ethan

[00:15:16.590] – Alex
It’s talked about like it’s everything is completely different.

[00:15:18.950] – Alex
It’s not you know, you’re still going to use, you know, if you have your AS advertised in your edge colocation provider. Guess what? It still uses BGP to advertise it outside. Say, if you have a rack of gear in there, it’s going to be the same rack that you would deploy elsewhere. Typically standard nineteen inch rack, standard power density, all this kind of stuff. You know, there’s various considerations you need to take care of, but it’s not it’s not magic.

[00:15:41.600] – Alex
You know, I think so much of the the press is it’s this magic new thing and it’s amazing and it’s really not, you know.

[00:15:48.290] – Ethan
OK, well you’re saying it’s something I think a lot of us have been wondering or thinking of. Like, isn’t edge just like you put a data center in a special place that you need it to be for the performance characteristics of the workload you need them to be. Is it really that magical of a thing? Because we have had edge consortiums and we had fog and we were talking before the show about mist Computing, all of these terms that are supposed to categorize some magical thing.

[00:16:11.840] – Ethan
But you’re saying there is no magic, Alex.

[00:16:14.150] – Alex
I mean, the magic is in terms of, to me, how you optimize the network connectivity, both on the front0haul piece to the access networks and on the backhaul piece to other data centers. Other larger IXs back in the market, in other markets to make the performance advantages of edge actually real. I mean, the rest of it should be and I think it should be standard as much as possible, because if you’re creating something from scratch, you should have a really good reason to do so, in my opinion.

[00:16:38.830] – Alex
I mean, especially something for potentially critical infrastructure. So, you know, in the ideal case, for me, I think an edge data center is basically a sized down version of a traditional data center with enough resiliency for the use cases that are going to be deployed there and a sensible set of network connectivity that lets you actually get lower latency than going back to the larger point in a market or an adjacent market. Apart from that, you know, from a facility design standpoint, you’ve got a few things like, hey, more focus on automation, more focus on unmanned operation and things like that.

[00:17:11.210] – Alex
But again, those things aren’t they’re not magic. You know, these are things that have been in process for a while. You can extend things that are occurring in larger data centers out to edge data centers, like more focus on automation, for example. But I just find it sometimes gets treated like it’s magic and you need edge networking. What is that? OK, it’s layer 2 eVPN between sites or dark fiber between sites, depending on what you own.

[00:17:36.890] – Alex
You know, BGP AS for the local set of edge data centers advertising their IP space out to the rest of the Internet. It all looks very similar to what we already do in many ways from an overall IT standpoint. So I don’t see the benefit of trying to trying to make something more complicated than it should be.

[00:17:56.240] – Ned
If I would find the magic anywhere, I would think it would be in the orchestration layer, because the big difference for me between centralized cloud and this distributed edge problem is now I’m not just running my compute in one or two regions, I’m running it in a whole bunch of little edge locations. And so I need whatever platform I’m using to be able to orchestrate actions across all of those little locations and provide the resiliency and the networking between them as well.

[00:18:24.590] – Ned
Like, how how do you think about that? How do you tackle that orchestration problem with the software that’s out there today?

[00:18:33.300] – Alex
That’s a really good point. You’re right, I do think there is a little bit of a divergence there purely because you’ve also got a couple of other dimensions, I suppose, to add onto that you’ve got cost and then you’ve also got physical location as well, adding on to what you were just saying. I do think that that will be a little bit of the secret sauce that the clouds are able to add to their own sort of edge cloud implementations. And that’s kind of where I think they’ll be able to differentiate a little bit between, you know, some of the the pure edge cloud plays that only have kind of perhaps more basic services and are focused on just one tier of infrastructure, like if you’re in AWS, for example.

[00:19:12.490] – Alex
So I can’t speak to anything they’re doing. But if I was them and I was looking at what I had in terms of data center resources, I’ve got honkin’ great big data center. I’ve got kind of medium, I’ve got potentially these local zone locations. And crucially, I know what my customer workload is because I’m providing the workload service typically. Right? It’s s3 or something else or Lambda or whatever it is. I have some intelligence as to what they need, what they’re willing to pay for, all those kinds of things.

[00:19:38.970] – Alex
So I do think there is a little bit of a gap there in terms of orchestration and I think particularly between what you’ll be able to do from kind of a closed cloud system. If you don’t, if I can kind of tar the public clouds with that brush, you know, there’s some parts that you don’t get to see, right. Compared to some of the more open source or available stuff like K8s that you might want to use. But did that answer to what you were thinking Ned, or is that just a ramble?

[00:20:03.580] – Ned
I don’t know. I I was waiting to see if you would bring up Kubernetes and not just Kubernetes, because obviously you’re dealing with physical hardware here. And depending on what how you’re consuming that platform, maybe you really just want empty racks where you can put your own gear. Now you need something that can orchestrate from the bare metal all the way up. So that’s not going to be K8s. That’s got to be some additional stuff on top of it.

[00:20:31.620] – Ned
Or if you’re going through one of these other cloud providers who is managing the site and the hardware for you, you might just be spinning up a microk8s cluster or a K3s cluster that’s going to provide the services. Is there anything that’s that’s. So anything out there that scratches that bare metal itch when it comes to fully provisioning and orchestrating a whole bunch of sites, because I can’t think of anything.

[00:20:56.800] – Alex
Yeah, I mean, I think the first thing that comes to mind would be something, again, kind of a, you know, a a vendor system, you know, like something like like the likes of Packet have, you know, for example, that’s kind of their secret sauce to orchestrate and set up bare metal. But in terms of something that the that the enterprise could just kind of take and use, I must admit I’m kind of drawing a blank on that.

[00:21:15.190] – Alex
Really, that might be because my my typical focus on this at the moment is kind of as a as a colocation provider. How how would this best be supported? Right. In that case, it’s kind of like, OK, there’s the assumption there that the colocation facility provides a lot of the kind of kind of the DCIM elements to inform the user potentially of is there power outages, is there this is there that, which kind of doesn’t solve the problem that you’re asking about.

[00:21:42.460] – Alex
But I kind of think it’s there’s that piece for the physical infrastructure. There’s the provisioning of the hardware from a kind of a. An IT use case stance, but I’m not saying this properly, but I hope you know what I mean. And then there’s the ongoing orchestration, I think, of the services that operate on top of that infrastructure. That’s kind of how I think about it. Does that match how you think about it or if I got this?

[00:22:04.640] – Ned
No, I think that’s that’s sort of the layer cake of things that needs to be provisioned to actually run whatever your cool new application is, whether it’s some crazy AR thing or gaming or finance or, I don’t know, health care, I feel like health care probably has something to play in there. Who knows? They are they are giant industry.

[00:22:21.710] – Ethan
[AD] We paused the episode for a bit of training talk training with CBT nuggets. If you’re a Day Two Cloud listener, you are you’re listening to it right now, then you’re probably the sort of person who likes to keep up your skills, as am I. Now, here’s the thing about Cloud as I’ve dug into it over the last few years. It’s the same as on Prem, but different the networking. It’s the same, but different due to all these operational constraints you don’t expect.

[00:22:45.800] – Ethan
And just when you have your favorite way to set up your cloud environment, the cloud provider changes things or offers a new service that makes you rethink what you’ve already built. So how do you keep up with this? Training. And this is an ad for a training company. So what do you think? I was going to say obviously training and not just because sponsors, CBT nuggets want your business, but also because training is how I’ve kept up with emerging technology over the decades.

[00:23:07.310] – Ethan
I believe in the power of smart instructors telling me all about the new tech, so that I can walk into a conference room as a consultant or project lead and confidently position a technology to business stakeholders and financial decision makers. So you want to be smarter about cloud CBT Nuggets has a lot of offerings for you, from absolute beginner material to courses covering AWS, Azure and Google cloud skills. Let’s say you want to go narrow on a specific topic. OK, well, there’s a two hour course on Azure security.

[00:23:37.190] – Ethan
Maybe you want to go big wide. All righty. There’s a forty two hour AWS certified SysOps administrator course and lots more cloud training offerings in the CBT Nuggets catalog. I gave you just a couple of examples to whet your appetite. In fact, CBT nuggets is adding forty hours of new content every week and they help you master your studies with available virtual labs and accountability coaching. Interested? Of course you are so satisfy your curious mind by visiting CBT nuggets, dotcom cloud and figure out if CBT nuggets will work for your training with their seven days free trial.

[00:24:13.490] – Ethan
Just go do it. CBT nuggets dotcom cloud for seven days free. That’s CBT nuggets dotcom cloud. And now back to the podcast I so rudely interrupted.[/AD] [00:24:26.220] This we’re talking about orchestration, Alex network and networking orchestration has lagged, certainly, and part of it’s just in this context, you’re talking about plumbing, the location network-wise being very key to to having it offer you the performance profile that you’re looking for. Well, there are services out there now that allow you to, depending on the colo that you’re in, dynamically stand up a link so you can improve that locality and latency, packet fabric is a company that comes to mind where you can via an API, basically call up a circuit.

[00:25:01.680] – Ethan
And as long as they’re plumbed into the colo where you live, you can pay on demand and have that be part of your orchestration. Is that something you’re maybe not Packet Fabric specifically, but just do you see that orchestrating network links as part of this?

[00:25:17.460] – Alex
I do. And I think I probably put MegaPort similar to similar plays in that kind of bucket, right. Where it’s that kind of API driven network fabric that you can you can call on programmatically based on various business challenges that you have at a particular time. I need an on ramp connection. I need this, I need that. I do think those are a key part of this. I think it’s interesting to again kind of separate the the network services that I want to be able to provide from the the raw underlying fiber connectivity between the sites.

[00:25:47.190] – Alex
Right. So I think you need to address both to really solve the challenge of whether or not an edge data center is going to be useful. You need to have the right fiber physically terminate in that facility. And then you also need to have something like a Packet fabric to provide the customer with the ability to provision the right connectivity from their deployment in that location to.

[00:26:06.540] – Ethan
Which is, so between that and and BGP routing. If you’re using Internet as transport, it’s not immediately obvious that what is geographically closer is closer. I live in the northeast US. Chicago, which is about a thousand miles away from me, is actually closer by some definition than if I were to put a server and stand it up in a New York City data center just because of the vagaries of how the fiber is laid out coming from where I am to where I’m trying to go routing and so on. And so being able to stand up these links, it really does assume a lot about the physical fiber that underlies the connectivity between the two sites.

[00:26:46.630] – Alex
Yeah, and that’s that’s something that’s been really interesting to dig into over the years with the kind of general theory around the edge data center right as in as we’re talking about a minute ago, if you just put it in the car park next to your house, that doesn’t necessarily mean your traffic is going to go right from there, right from your house to the data center and back. It’s most likely going to loop back around. And when you actually look at how the fibers is laid, not just in the US but globally, it doesn’t always go the way that you think it would.

[00:27:15.100] – Alex
In the US you know, a lot of it still, if you go sort of east to west, follows a lot of the old train lines, right? Yes. A lot of it emanates in like Denver. And you think, OK, well, that’s interesting. A lot of it is influenced by where the submarine cables land and then, you know, kind of hubs like like you mentioned, Chicago. So, yeah, I agree. You going to have both elements in there.

[00:27:36.010] – Alex
And I think that some of the some of the challenges that I’ve seen over the years with kind of edge data center is some people think that you only need, you know, the Packet fabrics or the MegaPorts, or the API driven kind of connectivity of the world. But unless you have them riding on top of the right physical connectivity between the locations that you want, yes, you may get bits delivered between the two locations and they may say yes it’s all going, it’s fine, it’s great.

[00:28:01.120] – Alex
But you need to know what path that’s actually taking. If you’re running a latency sensitive use case that has driven you to look at edge cloud in the first place. Right.

[00:28:09.490] – Ethan
OK. All right. So we’ve spent half of what we typically spend at a Day Two Cloud talking about definitions and what does a thing mean and beaten up edge cloud and so on. So let’s get let’s get practical. Alex, if I’m an enterprise, why what are the use cases where I’m going to build my own edge cloud? That’s that’s the thing that makes sense. Describe that for me.

[00:28:32.440] – Alex
Yeah. So a lot of these cases that I see that people are really going for are driven by either the need to process or store very large amounts of data in a very inexpensive or performance efficient fashion, if that makes sense. So the example that I tend to look at is something like a manufacturing facility or other location that’s going to generate, for instance, Ned you mentioned health care early to early an early today. I was about three minutes ago.

[00:29:00.100] – Alex
I thought, oh, but you know, something like that that generates a lot of data that can benefit from being processed very quickly is potentially very expensive to move over a large distance and may also have some kind of locality challenge as well. Like, for instance, a lot of health care information, for example, I think particularly over time, is going to get more and more local in nature as individual localities, especially in the US, have different data sovereignty guidelines for different types of information at the state level or even the city level in some cases.

[00:29:30.160] – Alex
Right. So that kind of gets to general category of inferencing, I think is a good candidate for an edge cloud use case where you might not have the space or the expertise in running something like an Nvidia EGX deployment or a DGX box that lets you just crunch through a lot of image processing on a on a an MRI scan, for example. You don’t have the in house IT to do that in the building. So you want to do it somewhere close enough.

[00:30:01.960] – Alex
That is still going to meet your performance goals. And then you can shuttle the data back and forth over short haul fiber, probably in the same metro area. That is the general use case and various sort of variations on that that my brain tends to go to. But there’s other ones, such as if you’re a network operator, actually building out some of the 5G network infrastructure that requires various latency targets between the RAN and the core network. But for the enterprise, I do think it’s mostly around dense processing of large amounts of data, which kind of to me generally goes back to inferencing.

[00:30:36.010] – Alex
Again, it’s not really well designed for long term storage, I think in many cases, because it might it might work as a hand off point for the cloud provider to then suck it up and take it back to a core location. But a 200 kilowatt data center is not going to be the best location for giant bulk data storage just because of the physical space it’s got available. Right. Does that make sense? Are you guys seeing anything different?

[00:30:59.740] – Ethan
Well, one thing implied in what you said is I don’t just build this for fun. I, I am going to build an edge cloud because I have a specific situation that I’m not going to be able to solve any other way. So I’m going to do this. It’s not like a new way to do data centers. And, hey, we’re going to put a bunch of these little ones and scatter them around is not it’s more like I have a very specific IT problem I need to solve and this is a way to solve it.

[00:31:27.280] – Alex
Yeah, that that’s really good. That’s a really good way to frame it. Thank you for that. That that’s how I think about it, because every physical piece of equipment you buy, every physical facility you’re responsible for operating and monitoring, it’s additional. Again, operational complexity you’ve got to deal with. It’s another check you’ve got to write. You know what makes this better than doing it in an existing data center that you have or using an existing cloud or an existing colo provider, it has to provide something that you’re not already getting.

[00:31:53.060] – Alex
Right. That’s valuable enough to warrant you incurring that extra cost complexity and kind of just to just go back something that we were talking about earlier, should an enterprise build their own edge cloud? Frankly, I think unless you are one of the very largest enterprises out there, like a large bank or something like that, I really don’t think it’s worth it. I would personally point people more towards a colocation, limited deployment in a edge facility or just the use of an edge cloud service and kind of burst to that as and when you need.

[00:32:23.780] – Ethan
Would you differentiate building your own edge cloud versus building some kind of edge infrastructure that maybe isn’t cloudy as such?

[00:32:31.310] – Alex
That’s an interesting, interesting point. I think it really comes back to really the value of the use case. Right. And if the use case that you’re trying to serve is and let me think about that for a second, because what are you thinking of? Some kind of like purpose built data collection rack or something?

[00:32:49.160] – Ethan
If it’s not cloud and I’ve got some operational model that I’m used to and I just want to throw a few racks of servers out there to host a bunch of compute for me. And I know how to do it. And it’s fairly inexpensive, relatively speaking. I could do it without it being too big of a lift on my IT team. That’s different from putting a let’s call it a cloud enabled stack or an automation enabled stack together.

[00:33:15.260] – Alex
That’s fair. I think in that case, you’ve still got the basic decision between is it worth it to me to put this in a physical edge location where I’m going to be constrained in terms of my price per kilowatt I’m able to get or am I flexible enough on performance that I can just deploy that in the existing colocation facility, for example, that I’m already using? I think in many ways it’s a very similar question whether or not it’s cloud or not like the cloud, but might change the complexity of deployment and then potentially the scale that you might be looking to go to.

[00:33:48.920] – Alex
But I think you’ve still got that that basic decision of, you know, if you need to use edge, I think you probably know you need to use it. And if you don’t need to use it, you probably shouldn’t.

[00:33:59.150] – Ethan
I mean, my gut is just from working at a bunch of data center environments as I don’t want another location unless there’s just some reason I’ve got to have one. I don’t need another location to think about. I don’t want to have to think about the network interconnectivity. I don’t want to have to think about what broke in there today. I don’t want to have to think about remote hands, etc., unless there’s really one compelling business reason to do it. That’s my instinct. So, yes, to me it’s not something I would take on lightly at all.

[00:34:27.440] – Alex
Yeah, I think you phrased it better than I did, and it’s kind of the how many throats, yeah? I mean I’ve only got two hands. So, you know, if I’ve already got my in-house IT, I might have a colocation provider. I might have a cloud already. Do I need to add an edge cloud? Can I do it through? If I need to have an edge location, can I do that either through my existing colo provider where I can still call the same guy up and yell at him. (Please do not yell at your data center support staff. They try very hard.) Or you know, my existing cloud if they’ve got a deployment somewhere else, like that’s kind of a middle ground I think where you’ve got a little bit of the flexibility of being able to deploy in an edge location, but you’ve still got some decrease in your operational complexity versus doing it self. If that makes sense.

[00:35:12.260] – Ethan
I would my instinct would be to try to solve the kind of problems that Edge is solving with better connectivity into my existing facilities. Is there and IX in town that I can use to to to solve this problem, get whoever the customer is I’m trying to talk to closer to me, et cetera. That’s that’s my gut. And then it’s like, no, the only answer is you’ve got to stand up a thing in this city and make it look like this.

[00:35:35.570] – Ethan
Oh, gosh, that’s it just feels like a whole thing. But I don’t want to deal with if I don’t again, if I don’t have to.

[00:35:41.060] – Alex
Yeah. I think that’s a really good angle to it because there are many cases when you when you know, you’ve been in some of those large data center environments, you might have two hundred, three hundred networks coming into that facility. And would the one that is happened to be being used definitely be the best one? Probably not. Have you looked at it? Did you did someone in purchasing pick, you know, OK, this one’s the cheapest per bit.

[00:36:02.510] – Alex
You know, this is the speed we need. Can you take out a circuit for less than the price of potentially managing and operating your own edge cloud deployment and get twenty milliseconds, fifty milliseconds less because you’ll running up, you know, up to Minneapolis from Chicago, rather than kind of going around and, you know, taking a loop around, I think that’s a really good point.

[00:36:20.210] – Ethan
So I would get twenty or fifty milliseconds back. That’s a lot. But anyway.

[00:36:23.830] – Ned
Yeah but that for me that brings up an interesting point, because one thing we haven’t discussed is the advent of 5G and all of the cellular providers that are out there because we keep talking about this big trombone effect and linking all these things together. And the obvious solution to me for that is local break out of traffic, being able to when it hits that Cell tower go, oh, well, this just goes to the next tower over, so I’m not going to, like, trombone it.

[00:36:51.830] – Ned
Is that something we’re going to see more of from from 5G and the cellular companies?

[00:36:58.820] – Alex
So I’m grinning at Ethan because I’ve been I’ve been waiting to suggest a show on this specific topic as I speak on this. But I can give you I can give you kind of that kind of the [inaudible 00:37:08] is I think the technology exists to do that. However, the cost and complexity of managing an exponentially larger number of network breakout points is really a difficult decision at the moment for the mobile operator to make. Right. If you think about what they actually have to do, they don’t get that for free.

[00:37:28.440] – Alex
Right. So you have in 5G, you got the distinction between the 5G, new radio RAN and then the 5G core. Right. And it’s by definition, it’s a very software defined distributed core network, but it still has to run on physical core network hardware that you put somewhere. Right. Right. So if I’m that network operator, I’ve got a choice between how how economical is it for me to deploy local break out in 10 locations across the US in large data centers that I’m already inter-connecting in versus do I do it in a hundred thousand, 10000 locations out more towards these edge locations where the network infrastructure isn’t there at the moment for interconnection?

[00:38:09.310] – Alex
So who am I going to bring onto that network? Who’s who am I going to pull my traffic off to to save me money on backhaul? And do I have the capability or the willingness at the moment to deploy and manage each one of those individual user plane function packet session anchor instances throughout the country? Right. So I think it’s going to be a gradiated thing. I think you’ll see some deployments where it’s like, hey, we’re going to go more regional than we currently are, where, you know, some of the providers may only have, you know, single digit kind of breakout locations at the moment just because of how the 4G networks were run.

[00:38:40.610] – Alex
So you could see that you could go to maybe 10 or 20 on kind of a regional basis. But going down to finer granularity, it’s definitely possible. I think it’s really going to be a change over time to see how far they actually take that out to the edge of that.

[00:38:57.070] – Ethan
Is it going to be a customer driven thing? They got some big customer that says, I really want you to build between here and here for me.

[00:39:03.650] – Alex
Yeah, I think I think that’s a good way to look at it. I think it’s going to be very customer and partner driven. Like, let’s say you’re a mobile operator who’s a partner with a cloud. If I’m, you know, Cloud A and I’m working with service provider B, I might say to them, well, look, you know, I’ll deploy this local zone equivalent in your little edge data center out by the tower. Then it becomes worth doing breakout in that location, because I’ve got the ability to pull traffic directly on and off of that cloud at the tower.

[00:39:29.240] – Alex
Right. Which is doing it kind of by themselves and waiting for it to come. I’m dubious.

[00:39:34.610] – Ned
Yeah. Verizon has a partnership with Microsoft to do private 5G mobile edge, but I think that also is going to be part of a larger push to just get Microsoft Azure gear in those towers and then that’ll create the basis for this sort of distributed cloud that Microsoft is trying to envision. So I think you’re right, the partner driven or customer driven is where we’re going to see it. I’ve never really trust telcos to innovate on their own.

[00:40:02.360] – Alex
Well, I would say, as I don’t I don’t trust or expect people to spend money for things that they don’t see a return from yet. Maybe I’ve been in the corporate world too long, but that’s that’s how I tend to look at it. Right. It’s like if I’m if I’m that service provider. Yeah, I could say that, hey, this enables edge. OK, well, what is the percentage of revenue that I’m really going to get from enabling in these 10 locations by me deploying and operating this local break out in my 5G core network in these 10 locations in St. Louis, for example? If that doesn’t pay me back, I ain’t going to do it right? Now as much as we would like them to like. I could read the standards all day long and I’m saying just put it here, put the packet session anchor here, please. Yeah, somebody’s going to write a check for it.

[00:40:48.020] – Ned
Right. Right. But they have the location. So if they can lease some of that location out to a partner and have the partner do all the heavy lifting. Now they’re making money by leasing their location without having to do all that technology stuff.

[00:40:59.650] – Alex
There you go. There you go

[00:41:02.150] – Ned
Now we’re thinking. Another thing that I’ve been thinking through as a thought experiment. And I’m curious to get your thoughts on it. From an edge cloud perspective. We tend to have this view of all the data flowing up to a centralized cloud, like, yeah, we’ll do some processing, but really it’s all got to flow to the central cloud. Is that is that a flawed way of thinking? Is it possible to just keep everything distributed at the edge cloud layer? Is that a model that would make sense to you?

[00:41:29.300] – Alex
Yeah, I’m going to go back to my my typical. It’s become my new. It depends. It depends on the use case. Right. Because I think if you if you think about kind of the inferencing example, I mean, let’s let’s pick one of the common things that gets mentioned, let’s say collecting data from autonomous cars. Driving around the city, right, let’s say you’ve got a bunch of 5G connected cars, all spooling data off onto the network, and then really, how much of that do you need to process?

[00:41:55.980] – Alex
See if the sensors gone haywire versus how much do you really need to send overnight to the core data center? It starts to look very sort of 80/20, 90/10 at that point. I do think there’s a lot of use cases that you point out where I frankly don’t care about the data once it’s been processed or I only care about the result of the inference. I don’t so much care about the raw data. So in those cases, I think the usefulness of, not the usefulness of but kind of the percentage of data that’s sent up to those core data centers is drastically reduced versus something where I’m going to collect every piece of data and store it forever.

[00:42:33.930] – Alex
You know, if that makes if that answer makes sense, I think it’s very use case dependent. And I think there’s a lot of cases where the mass amount of data does not really leave the edge cloud. Can you just transmit any any errors, erroneous readings, you know, things you want to do, further training on or learning on up to the core data center, and then you pretty much discard the rest of it after a period of time at the edge, because that to me has always been one of the potential advantages of the model.

[00:42:58.910] – Alex
Right. If these are just kind of distributed, you know, catching locations for all the data, not as useful as if they can act on the data and your application can determine what data you actually need to store because, of course, you’re paying for all of this. Right. So that did answer what you what you were thinking.

[00:43:16.130] – Ned
Yeah. Yeah. That’s kind of where my mind is that as well. And in the long term, I think maybe you could see a step away from centralized data centers. But there’s always that sometimes there is a point to centralizing data for specific tasks.

[00:43:30.350] – Alex
So, yeah. Yeah, I think there’s I don’t think and this is kind of we were talking earlier about the way that I originally kind of looked at edge data center market in the way that I now look at it. Like I kind of looked at it first as all of these applications are going to be kind of driven from data being generated at the edge. So it’s kind of an edge to core workload model. Now, I kind of look at it as well, but the vast majority of things we’ve figured out way to do, figuring out ways to do things in the core data center, in the cloud, and then we will pull elements of that that need real time response or real time processing out to the edge.

[00:44:02.960] – Alex
So I kind of look at it in that way that I do think, as you said, there’s always an advantage to having that centralized point, even if it’s for coordination, you know, ultimate data storage, workload management, and kind of treat each one of the edge nodes as kind of tethered off of that. That’s kind of how I think about it at the moment, rather than like a fully autonomous kind of edge cloud by itself that then has kind of a bidirectional relationship to a core cloud, if you will.

[00:44:27.830] – Alex
That makes sense.

[00:44:28.880] – Ned

[00:44:31.050] – Ethan
Is the core cloud running Kubernetes, Alex, is that we have to talk about it. It’s kubernetes, it’s a cloud show. We have to talk about. So is that kind of the common denominator or are we is there a broader scope that we’re going to be considering?

[00:44:49.120] – Alex
I think I think I look at either Kubernetes or any sort of cloud specific scheduling or management tool if my my public cloud demands that, I use that. But there’s no real reason, in my opinion, a lot of time to diverge. I’d probably just irritate many thousands of people. Usually my usually know a great reason I find in these use cases to diverge really significantly from kubernetes. I think it should be treated as an extension of what you’re already doing in your data center deployment and then hopefully build some location and cost awareness onto that over time to manage the, you know, the edge parts of that deployment.

[00:45:26.070] – Alex
If that answered what you would thinking Ethan.

[00:45:27.990] – Ethan
Yeah, yeah. You’re getting to it. It’s we’re seeing it is the kind of the common denominator kubernetes is either there right in your face and you’re interacting with it directly or there’s some other layer on top of it. But it’s kubernetes underneath. It doesn’t. I can’t tell if we’re still in the hype cycle, and that’s why we’re seeing it everywhere and people are going to like wear out of it, just kind of get tired of kubernetes and move on to something else, or if it really is going to be part and parcel of how we deliver applications for the long term.

[00:45:59.140] – Ethan
I’m not convinced either way yet because it really does seem like overkill. Like like for me, kubernetes, I fire it up and it’s it’s a Rube Goldberg machine to accomplish a simple thing like you know, slice a vegetable when I could have just used a knife to slice the vegetable. Why did I why do I need all of this stuff? Sometimes I wonder. But of course there are valid answers to that question and it does seem the common way people are going.

[00:46:24.370] – Alex
Yeah, I personally think it’s likely the way that this is going to keep going, because I think a lot of the technologies that we now look at as having one defined use that we will understand how to use kind of started off as that more kind of tool kit approach. I mean, if you really you know, you’ve see this Ethan, if you read through a lot of the BGP specifications, you can use that for a lot of things. It washes dishes.

[00:46:45.880] – Ethan
You certainly can.

[00:46:47.240] – Alex
You know, you can use it like as a datastore. You can do all this kind of just like but nobody uses it for that because over time we’re just going, OK, this is what you actually need. And it kind of funnels down to click this button, do this, run it this way and then fine. And a lot of the other kind of tools in the toolkit get kind of left behind because they’re extraneous or that’s what I that’s what I personally see happening with this. But.

[00:47:09.340] – Ethan
Another question for you, Alex, near the top of the show we were talking about edge, and you made the point. It’s not really that special. It’s just kind of a different flavor of what we’ve been doing for a long time. But there’s all this hype in the industry around Edge such that there have been these consortiums and standards bodies and so on around that term. For folks that are considering a deployment of edge cloud, is there a consortium or standards group or something that they should be paying attention to?

[00:47:34.000] – Alex
Yeah. So the best one that I’ve run into over the years of fiddling with this stuff is the LF Edge the Linux Foundation’s Edge group. There’s a lot of good projects in there, I think Akraino’s part of that. I believe a bunch of the other projects that that we’ve we’ve talked about in previous shows have done like glossary of edge computing I contributed to. A bunch of the things about how. Do we even need a new technology to address the edge use case? Or can we utilize something else and kind of manipulate it to do what we needed to do?

[00:48:05.320] – Alex
I’ve always had pretty good conversations with all the different projects on the LF Edge side of things. So that’s where I would I would typically point people towards. I think there are a lot of industry consortiums and single vendor things from the like. So I get emails all the time about them and things that kind of look like consortiums, but kind of aren’t when you look close enough and they kind of just talking about the device side of edge computing or a whole vendor’s kind of portfolio, but not bringing in other voices or kind of projects into it. So for my money, I think LF Edge is the best place for that.

[00:48:37.780] – Ethan
Well, it’s good to know that because it’s good to know that there’s some group that’s not a waste of time or just trying to ride a trend or something like that. So Linux Foundation Edge is a group worth paying attention to. Very good, Alex. We’ve talked about a lot of stuff over the last forty five minutes. This is just felt like, you know boom, a lot of rapid fire information. Pick out some some takeaways, some highlights from our conversation that you want to leave people with.

[00:49:01.150] – Alex
Yeah. So I would I would kind of point back to a couple of things that I talk about in in my book, which is Understanding Infrastructure Edge Computing, full title is, Understanding Infrastructure Edge Computing: Concepts, technologies and considerations. There you go.

[00:49:13.570] – Ethan
My goodness.

[00:49:14.500] – Alex
But I had to make the subtitle longer than the actual title. But I would say, you know, there’s there’s still not a great solid definition of what edge or edge cloud is. I would say identify exactly what you and the people that you’re thinking about spending money with or asking for money for if you’re trying to buy something for your IT business, really are talking about.

[00:49:36.230] – Alex
I would say that a good place to start is in the book. A lot of discussion is about the different tiers of edge data center and related infrastructure that they have need to have to be useful. I would say that treat edge and edge cloud, I think as commonly as possible, if that makes sense, ideally as an extension of your existing infrastructure choices that you’ve made to avoid being a super sacred cow that nobody’s nobody’s allowed to touch and ends up needing special care and feeding.

[00:50:05.080] – Alex
And it’s a whole you know, it’s a whole big thing that becomes more trouble than it’s worth. And then I think I would also, just as a last point point to especially with edge cloud or really any edge data center deployment, pay good attention to the network topology and specifically the interconnection locations for the networks that you actually need, especially on the access side, which is not always easy to find. Right. You can’t just go and ask a cell provider, hey, where do you break breakout?

[00:50:31.270] – Alex
They won’t typically respond to you very kindly. But, you know, if you’re talking to an edge cloud provider or an edge colocation provider or someone like that or you’re thinking of, hey, maybe this would be a good location for me to do this myself, definitely. Pay attention to that, because the network piece is what’s really going to determine whether or not you’ll meet the performance benefits of your edge deployment and if you’re not going to meet them, why are you doing it?

[00:50:55.840] – Alex
So that’s that’s my that’s my spiel.

[00:50:58.600] – Ethan
Very good. Thank you for all of that. All of your highlights, all your thoughts here. And that book, again, if you’re listening and interested in this understanding infrastructure edge computing, I just went up to Amazon and typed in Alex’s name, Alex Marcham, and that was the first book that popped up right there at the top of the Amazon list, along with other books that Alex has written and are available, if you are so interested. Alex, final question to you. Do you have a blog, Twitter handle? How can people follow you on the Internet?

[00:51:26.590] – Alex
Oh, I need to keep my blog up to date. I haven’t done for a little while now, so I won’t I won’t chain myself by revealing the address. But I am typically found on LinkedIn these days. I’ve managed to swear off almost all other social media, but I’m still a very still very nice person, I promise. So just get in touch on LinkedIn if you want. And I’m happy to chat about anything.

[00:51:45.010] – Ethan
And you’re defensive in defense of your blog. I mean, dude if you’re writing books, how do you have time to blog? Right. So that’s. Come on, man. Well, Alex, thank you very much for appearing on Day Two Cloud today, and Virtual High fives to you for tuning in. If you have suggestions for future shows, Ned and I would love to hear them hit either of us up on Twitter at Day Two Cloud show or fill out the form on Ned’s fancy website, Ned in the cloud dot com.

[00:52:09.460] – Ethan
And if you’d like to chat with other IT Ops folks living the day in and day out technology life, join our community Slack group at packet pusher’s dot net slash slack. Read our three simple rules about how to be a good slack member and then sign up. Everyone in the community is welcome. Even even vendors just obey the rules. Until then, just remember cloud is what happens while IT is making other plans.

More from this show

Day Two Cloud 174: Building Kubernetes Clusters

On today's Day Two Cloud podcast we walk through how to build a Kubernetes cluster to support a container-based application. We cover issues such as what constitutes a minimum viable cluster, rolling your own vs. Kubernetes-as-a-service, managing multiple...

Episode 102