Microsoft has experienced a renaissance. By everything that we've seen coming out of Microsoft over the past few years, it feels like the company is really walking the walk. Instead of just talking about how it’s innovative, it’s demonstrating that. Microsoft has been on an amazing journey, making the progression from telling customers what they need to listening to them and responding by building what they ask for.
Today, we’re talking to Corey Sanders, Corporate Vice President of Azure Compute at Microsoft.
Some of the highlights of the show include:
Customers are asking for Microsoft to help them through support and enabling platforms
Storytelling efforts through advocates, who play a double role – engaging and defending Microsoft
Customers moving to the Cloud are focused on a continuum and progression; they have stuff to move from one location to another and want all the benefits–better agility, faster startup time, etc.
Virtual serial console into existing VMs; this is how people are using this and Microsoft is going to, if not encourage this behavior, at least support it
Microsoft is the only Cloud with a single-instance SLA
Serial consoles: Windows' has seen less usage, partly due to operational aspects of Windows vs. Linux. It's not a GUI; it's scripting.
Does the operating system matter? From a Cloud perspective, it shouldn't have to matter; you should be able to deploy it the way you want
Edge enables much more complex and segregated scenarios; that combination with cognitive searches running locally will make it accessible anywhere
Branding challenge as customers start to notice that devices are smarter and more complex; will they lose awareness that Microsoft Azure is powering most of these things - they shouldn’t care
An awareness of not just what's possible, but what's coming; the democratization of AI
Education and fear gap of trying something new and taking that first step; make products and services stupid and simple to use
Customers return to add cognitive services and AI capabilities to existing, running deployments, environments, and applications
Multi-Cloud solutions can be successful, but there's a caveat; they’re actually built on a service-by-service perspective
Azure Stack, offers consistency, but some people may place blame on it for poor data center management practices; some expectations and regulations may be frustrating to some customers, but lets Microsoft offer a consistent experience
Freedom and flexibility have been challenges for Microsoft and other products for private Clouds
What people need to understand about Azure, including from a durability and reliability experience
To some extent, scale becomes a necessary prerequisite for some applications
Microsoft has taken many steps and is the leader in various areas
Corey Sanders on Twitter
Full Episode Transcript:
Corey Quinn: This week's episode of Screaming in the Cloud is sponsored by ReactiveOps, solving the world's problems by pouring Kubernetes on them. If you're interested in working for a company that's fully remote and is staffed by clued people or you have challenges handling Kubernetes in your environment because it's new, different, and frankly, not your company's core competency, then reach out to ReactiveOps at reactiveops.com.
Hello and welcome to Screaming in the Cloud. I'm Corey Quinn. I'm joined this week by Corey Sanders who is excellently named. He was the head of product at Microsoft Azure Compute and he's now the Corporate Vice President of that same group, meaning that everything that happens in Azure Compute is, in one way or another, your fault.
Corey Sanders: That's right both good and bad.
Corey Quinn: Perfect. We'll keep our slings and arrows here for a while. How long have you been with Microsoft?
Corey Sanders: Approaching 14 years, actually. I started right out of college and I started out as a developer. I was in the Windows Serviceability team and I fixed bugs in Windows in the early days, and then I moved over and then became a product manager for Azure. I was, I think, the fifth PM on the team so very early days and been on it since. Yes, it's been a lot of fun.
Corey Quinn: Wonderful. What has it been like over that time period? I mean, you go back to 14 years ago, that's 2004. That was after the days when Microsoft was in the world of, "Yes, this internet thing. That's not going to happen," and more into the days of, "Cloud? That's not really what we do either." In the last decade or decade and a half, there's been a tremendous renaissance. I think that Harvard Business School and the rest will be studying this for years as a fantastic transformation story of a company reinventing itself. Everyone claims to want to be able to do this but by everything that we've seen coming out of Microsoft over the past few years, it feels like you're really walking the walk. Instead of just talking about how you're innovative, you're demonstrating that. How does that work? How did you get to where you are from where you were?
Corey Sanders: It's been an amazing journey, a rollercoaster even, as we made this progression, and I do think a lot of it originated from the work of doing this in the cloud. I think the biggest thing is the rate at which we learn and listen from our customers changed, and that data, I think, it's probably the biggest thing I had some retrospective, thinking hard about, "How did this all happen?"
I've been in Azure now for close to nine years and, "How did this all happen?" The way in which we listen to our customers and respond is what, really, I think changed them. In the early days of Azure–and I just can't believe–it was very much, "We know what's right. We know the platform that you guys should be using and let us show you that platform." Over the years, it quickly changed into, "We actually may not know, inherently, what's right. We may need to actually build what people are asking for."
Really, put simply, that is where the springboard and sort of everything else came from. If you look at things like support for Linux, support for things like PostgreSQL and MySQL, support for many of the open platforms, many of the other solutions, even support for things like Edge and some of our intelligent platform, all of these really spin off from what our customers are asking for us to help them do, what our customers are asking for us to enable. I think that's really been the crux of that transformation and then the freedom, from a management perspective and a leadership perspective, to go do that.
Corey Quinn: From a customer story perspective, something that has been incredibly noticeable in the community for the last two or three years has been Microsoft more or less hiring every dev advocate, developer relations, developer-person that they can find, to tell stories. Is that storytelling aimed internally at Microsoft? Is it aimed externally to the community? Is it helping to build the story before it winds up getting out there?
Corey Sanders: I think the best advocates that we have are the ones that are actually looking both ways and their harsh-poked directions. What I mean by that is our best developer advocates are really mean to me about the things that I got wrong, the parts of my service that are no good, the things that people are struggling to use. Then, they're also pushing back hard in the community and saying, "Hey, you guys are being overly judgmental. You're being overly harsh." They're sort of playing this double role–sort of a double agent, if you will–at both engaging with us and yelling at me while also defending the platform externally. It's a tough job, actually. I think that I really respect a lot of those folks. It's a hard job.
Corey Quinn: It is definitely not for the faint of heart. It's also a bit of a challenge in that back when you first started seizing up developer advocates, I'm going to say Azure was not as great as it is today. Is that diplomatically acceptable to frame it that way?
Corey Sanders: Yes. I would like to say that in every day, that's true from the previous days.
Corey Quinn: Exactly. We are closer than ever, it's uplifting and meaningless all of the same time. It felt at the time that, "Okay, they're bringing in a lot of people to tell a story, but that story right now isn't great." I don't think people have that same criticism anymore so I'm wondering how much of that was, "Now that their story is better, they're bringing in more dev advocacy folks," versus, "They brought in the right dev advocacy folks and now, as a result, things are much better than they were."
Corey Sanders: It's a little bit of both. I think that the developer advocates definitely have a major impact in Windows. Especially, I think, the emphasis is–the developer advocates that we may not, inside Microsoft, have as much experience. The strongest pushback, the job and the known developer advocates come in and say, "Whoa, you guys completely missed the bone on this," that has been the biggest push for us because it's just a voice that we haven't heard as strongly.
Making sure that we're hearing that has been really important and definitely getting the right developer advocates has been a big part of it as well. There's more room to go. I think that if you sat down with the developer advocates, they'd tell you, "No, there's quite a bit still that needs to get fixed." I think we still got a long ways to go but we're definitely better and I think they have it reviewed again.
Corey Quinn: That's a very fair answer. Changing gears slightly, the keynote this morning and the ones that are upcoming over the rest of this week are focusing on a lot of the higher-level flashy wonderful services–and we will get there but 85% of global spend these days, as for several vendors, is not based on these high-level platform as a service offerings; they're based on compute, storage, maybe a managed database here and there, an object store, the data transfer between all of these things.
As you take a look at customers who are moving into the cloud, either existing on premise environment or looking to migrate from either another cloud provider or, in many cases, their startups who are, "All right, time to get started. We have this idea. Turns out: In 2018, building a datacenter is not near Top Five List of Things to Do. Let's go with a cloud provider." How do you see that from your side? Do you find that customers are interested and very all-in when it comes to these high-level AI, ML, edge-type of services or they say, "This is great and we want to talk about this but, first, we need a really big box to go ahead and crunch some numbers for us."
Corey Sanders: Absolutely. I have a skewed perspective, certainly, since I'm the Compute team so, certainly, the focus that I get is, "I need that gigantic box. Can I get that and can I get it now?" For me, the conversation of customers is almost always about a continuum. It's almost always about a progression. As you've stated, most customers today have a bunch of stuff that they're running on print, or running on the cloud or what-have-you, and their first order of business is moving that stuff from one location to another location, and getting all the benefits there–better agility, faster startup time, whatever–but the first step is just moving the stuff.
We have a bunch of services and tools to try and help that for exactly that reason. I think the key reason why you look at some of the keynote today and some of the future directions, it's really important because there's a lot of value there but nowhere near as much value when you look at the broader pass services, when you look at the new AI opportunities that exist, when you look at the opportunity to deploy things onto the edge.
There is so much more value, business opportunity and cost savings in those hired services that seeing the vision there is really important. When I look at the overall continuum for customers, they're going to be along this path. In some services, they're going to lift, and shift, and stop, and say, "Thank you very much. That's my day," but for some, they're going to say, "This is business-critical. This is where I want to go really innovate, and I'm going to carry that all the way through and I'm going to start looking into these new services, and capabilities, and things that I heard about it at Build," or other such conferences. I think that that continuum's important, but you're absolutely right. For a lot of customers, the first step is just getting into the cloud and then to take it from there.
Corey Quinn: To that end, you had a blog post about a month and a half ago about a feature that wasn't widely-recognized, but I saw that and I'm doing backflips, personally.
Corey Sanders: Four tweets and I've gone on anything else on that ever, by the way.
Corey Quinn: Absolutely. It's a great offering. It is a virtual serial console into existing VMs, and other providers have pushed back against this exact thing by saying, "You should be viewing cloud instances as ephemeral. They should be cattle, not pets," which is a very polite way of telling your customers to go screw themselves. It's easy to write a whiteboard diagram with the perfect architecture that makes sense but for what a customer should be doing, but they have to get there from where they are.
As much as we'd like to tell people, "Don't have any single instance be something that needs to be cared for lovingly," that's not the reality today so people had to do all kinds of unfortunate workarounds. This seemed almost like a tacit acknowledgement that, "This is how people are using this and we're going to, if not encourage this behavior, at least support it." Was that difficult to win hearts and minds around internally before release?
Corey Sanders: It's interesting. I think if you had asked me that eight years ago, the answer would have been yes. It would have been, "That's not our model. That's not how we do cloud." In the last year, it wasn't. I think the hardest part was implementing it in a secure way. I think it's actually just a complex thing. Even though the experience we produced is pretty simple, building it was a little bit harder than I thought it was going to be.
I think, to your point, it really helped establish my place in the world that, again, this was sort of this serial console–again, the basest of the base, sort of this core infrastructure access thing–got more excitement than any other announcement that I've ever done. It was very telling to me of the world in which our customers are living, and that's great. I think that there's no similar, single-instance to SLA. We're the only cloud with a single-instance SLA. It's a similar interesting thing where, to your point, customers have this today and they want to take that first step. How do you make sure that first step isn't totally painful and a huge step so that they can take the following steps after that?
Corey Quinn: In many of large accounts, you see customers who are proudly up on conference stages, talking about how they have no single points of failure, any instance can be terminated with no business impact, and then you go out for a drink afterwards and you pour six of them in when they're not looking and then they become really honest. "Yes, we have no single points of failure except that one," and that's always a database server or it's the Jenkin's Box that does CI. Yes, incredibly so, and if it goes down, everyone's having a terrible day.
I'm old enough to remember working in datacenters where I was thrilled to get serial consoles in because I no longer had to frantically drive to the datacenter at 3:00 in the morning to figure out what broke. Seeing this come to the cloud, it's, "Oh, yes!" I'm old enough to remember why that's important and, I guess, experienced enough to understand that, as much as we'd like to say this is the perfect design pattern, nothing is perfect in the real world. Everything's a tire-fire.
Corey Sanders: That's right, and what's funny about serial consoles is we've seen it in action. It's had a huge uptick on the Linux side, and this is one of the more interesting things, also, about our base, it's had a huge uptick on the Linux side, huge amount of users. The Windows' side we've actually seen less usage, and that's also partly some of the operational aspects of Windows versus Linux. It's not a GUI; it's scripting.
We've seen less usage on the Windows' side. The Linux side has actually been a huge amount of growth, huge amount of kick-up, and so we now also need to go back a little bit and think hard about our Windows' support here because I feel like we nailed it on the Linux side and gotten sort of the same response that you're evicting here. For the Windows' side, we've got more work to do so this is kind of interesting.
Corey Quinn: Again, I go back in time years ago and I would never have believed I'd be able to ask this question without being thrown out of a Microsoft event. In 2018, does the operating system matter? If you're using a server-less function that fires off a piece of code that you give it or we're talking to a database that's managed for you, you likely don't care about what operating system is under the hood. You care about the code that's executing and you care about the business outcome. How does Microsoft think about that given that they have, for a very long time, been the operating system company?
Corey Sanders: There's sort of two sides to this conversation. From a cloud perspective, an enabling scenario perspective, it shouldn't have to matter. It shouldn't have to. You should be able to deploy the language that you want, you should be able to deploy it on the platform you want, and you shouldn't have to care. This is especially when you look at those higher-end past services, things like functions, things like PostgreSQL, MySQL, Cassandra-based services, but it doesn't matter what's running underneath that thing. You're getting a service and you're taking advantage of it.
Even something like the intelligent engine, the IoT Edge solutions or you're running containers. It really shouldn't have to matter what's running them. It's enabling you to be able to deploy your services and solutions, very much so when you look at these platform services and solutions. Now, certainly, I think there is an aspect of, because of our long history with Windows, we do have a pretty strong belief that we run Windows very, very well.
Corey Quinn: If you can't, I think it's time to give up.
Corey Sanders: This is certainly an important factor. I think when we talk with one of those customers, we do strongly believe that aspects of both our licensing and our support model and so on, obviously, Windows runs incredibly well, again, for those customers in that first phase. When you look at that later application platform, that edge solution, it shouldn't matter, and that's really how we design our solutions.
Corey Quinn: Today, there was a lot of talk about being compute being done at the edge, a lot of IoT stories, a lot about machine learning/AI/math if it's just two engineers talking to one another and not trying to raise money. How do you see that evolving in driving the growth of Azure as it continues to embrace a world-spanning computer as Satino Dallas said this morning.
Corey Sanders: I think we have a really unique perspective on how the cloud has evolved. Certainly, we have the most global cloud of many of the clouds, with 50 nascent points and a comprehensive set of platforms, but the realization in understanding that, as more devices and more computations required of the edge, whether it be small devices like Raspberry Pi's, Raspberry spheres, or whether it be large computational devices to something like an Azure's pack enable you to take the same cloud model, the same developer experiences, the same containers, this is why we've really centralized our focus on containers being this portable object that could deploy anywhere, taking those same ones, building, creating them in the public cloud, developing them in the public cloud and then sending them out into the edge build to run the computer as close to the end-user as possible.
I think it's going to enable us some scenarios that we've only dreamed of. In fact, the demos on stage, I think, were really exciting, like taking a camera that doesn't need to communicate back at all to be able to do cognitive circles, to be able to do visual recognition and fire a function based on that cognitive service, all without leaving the device is astounding. It's something that I think is so cool.
With the edge not necessarily being 24/7 connectivity in the public cloud, and still enabling it to run. I think it enables much more complex scenarios, much more segregated scenarios, again, from top to bottom. I think it's unique, I think it's very exciting and I think the combination of that with the cognitive searches running locally is going to make the edge accessible to homes anywhere.
Corey Quinn: Do you think there's going to be a branding challenge as people start to notice that devices around them, from doorbells, to cameras, to wristwatches, et cetera, are smarter and able to do incredible levels of complexity? Do you think that people are starting to wander away from the idea of that even being tied to a cloud provider at all? It's not – for example, I was talking on the show earlier with iRobot and some of the stuff that they're working on.
People don't think in terms of their provider and the services they're using; they just think of it as a vacuum that no longer hurls itself down the stairs to its own death. Do you think that people are going to lose awareness of the fact that it is Microsoft Azure technology powering most of these things other than the folks who are building it? And, if so, is that a bad thing?
Corey Sanders: I've always found that one of the more magical aspects of technology is people don't need to care. Whether you're end-customers that are using these devices, it doesn't matter. They're getting the value that they're getting from whoever they bought that device from. Being a platform company and offering this technology, it's not about jamming ourselves into people's lives; it's about helping people's lives without having them to care.
I think that the future of these devices and customers interacting with them is going to be exactly as you say: Everything is just going to get smart and people are just going to start taking that for granted, which is an amazing, amazing future because, suddenly, you take that for granted and that's just the way things work then, but it doesn't necessarily matter to me whether people then say, "Thank god Azure's there." Being a platform company is fine. We're there to support that service, and that end-customer, and that advice and hopefully make it better and easier to build, and make better, and better, and better, but I don't think that there's a branding risk there and I think that, actually, the benefit of technology is that they don't have to care.
Corey Quinn: One of the most striking moments from the keynote this morning was AI for accessibility, tie idea that this technology isn't just about ephemeral business outcomes and, yes, we see this beautiful chart that's generated for us automatically, great. That adds value but it doesn't make people sit up and take notice the same way of, "I'm blind, my child is not, but now I can work with them on their homework."
That is one of those gripping, very compelling sentimental stories that, I think, shocks people into an awareness of not just what's possible but what's coming. If you had asked me about that three days ago, I would have said, "Oh, yes. That'll probably come in a couple of years. It's here today," and that's something that's incredible to wrap your head around. It's got to be an interesting experience, watching the future arrive rather than always being this thing that's one day coming and maybe your great grandkids will have a flying car.
Corey Sanders: What a heart-wrenching video that was, too. It's touching. When I look at that, the thing that starts to talk a lot about is the democratization of AI. It's sort of, "How do you get AI into everyone and anyone?" Then, when you look at those types of services, it drives it home. I think that the demo that Jeff did today where he takes a picture of Scott and gives a checkbox, it's cute and it's a great demo to show off the power of AI but it doesn't really drive home why democratization matters, why giving people who understand accessibility the ability to take advantage of AI to improve the lives of hundreds and thousands in excess of billion people in this world fundamentally change their lives. I think that, again, is a touching opportunity for technology, and I think it really drives home where AI is taking the world. It's a really exciting opportunity.
Corey Quinn: I wound up conducting a survey somewhat recently and, as a result of it, I started using the term AI and machine-learning lesson last because I was asking people, "Is this something that folks are actually using or is this one of those far future pipedream things that, yes, a couple of people are doing interesting things with and duping VCs out of giant piles of money but it's not here yet?"
And the responses were fascinating not just because of the breadth and depth of responses–and they were all across the board of what people were working on–but the fact that they almost universally started with some form of the statement, "Well, I'm not a data scientist but," it's something that people feel like, " I just have this silly thing that isn't taking advantage of any of that but here's what it does," and they're wrong.
What they're doing is fantastic. They have effectively become someone who is capable of wielding these tools but there is an education and fear gap of, "Well, I'm nowhere good enough for that." I fought for 10 years against being called a developer despite the fact that I spent 80% of my time writing code so it was a difficult mental barrier to get over. I think that we're going to see less of that with technologies like this as they continue to evolve and become used for interesting world-changing technology applications like we've just seen, but it is interesting to me. How do you get people to take that first step? How do you get people onboard-ed into a, "Yes, we machine-learn and you can, too."
Corey Sanders: Just like with anything with developing, you make it stupid and simple to use, like, "This is where you look at our cognitive services," and the ability to just take resting points, API endpoints, train them yourselves by just uploading a set of images and then being able to just dump that into a container and run it. They're sort of like, "Why aren't we all using AI?" I think that there's an aspect that's just so simple to take advantage in this form.
Sure, we have the very complex, massive amounts of GPU and FPGAs and so on and so forth and that all is there, too, but to your point–"I'm not a data scientist"–and so I'm excited to go just write a logic app, write a function that's going to do a video analysis and be on with my day.
Corey Quinn: I like that. It's going to be very difficult to pronounce when reading it but it works better when spoken.
Corey Sanders: People will probably misunderstand. We don't need to get into that right now.
Corey Quinn: That'll be an after-show discussion.
Corey Sanders: We can spend some time on that over some cocktails.
Corey Quinn: Absolutely. As far as people who are adapting this and using this, are they generally–in the abstract; I don't need specific NDA-violating, business-ending stories here–but do you find that people who are adapting this and working this are already as your customers or are they coming in with, "We use either other cloud providers," or, "We're from the stone age and run around a datacenter after we mine our own TIN to build the server." I mean, where are you customers coming from?
Corey Sanders: They're using visual cognitive services to find the right TINs to go online.
Corey Quinn: Wonderful.
Corey Sanders: I think I'll say a lot of them are coming in as already customers. Going back to a little bit of the previous conversation, I think a lot of them are coming in as, "My first step is migrating my current stuff to my new stuff and then I'm going to start adding some services to it," and so we see a lot of the customers coming in as adding cognitive services and adding AI capabilities to existing, running deployments, environments and applications.
It becomes some of the easiest way, again, coming back to a little bit of sort of accessibility of just being able to use it, and the easiest way to use it is to append to an existing app so we see a lot of that. There is some as we're seeing more and more of server lists and multi-cloud server lists picked up, which has been kind of an exciting trend that we've started to see pick up. There's a little bit more of taking advantage of our cognitive services in conjunction with other cloud-based server-less products and bringing them all together. I think that's also a very exciting multi-cloud opportunity, but a lot of it is sort of adding, and building, and supporting upon existing services and solutions.
Corey Quinn: Do you see multi-cloud as being a driver these days? I'm somewhat bearish on the concept myself. I find that when companies have specific workloads that they want to be able to deploy to multiple providers, it invariably turns into a world where they're no longer taking advantage of any higher-level service whatsoever. At that point, they're running VMs with storage, maybe a database and that's as far as it goes so they spend a lot of time reinventing things that they could get "for free" if they wound up picking a single provider to work with for that workload.
Corey Sanders: I see a lot of successful multi-cloud solutions but there's a caveat. The successful multi-cloud solutions I see are ones that don't fall under your trap, where they actually built it on a service-by-service perspective. They'll take one service here and one service here, and they will take advantage of all of the deep platform capabilities on each side. This is where capabilities like Azure Kubernetes Service or the Kubernetes platform being able to span across multiple solutions offers as flexibility, as portability, while still taking deep advantage of past services, even causes EV, multi-model data solution, you can write Cassandra-based workloads here, a fully managed platform; you can move them over here and still have your Cassandra work seamlessly. This is where I do see success stories, but not limiting yourself to the lowest platform but finding portable platforms that can go up the stack.
Corey Quinn: Wonderful. Data gravity, of course, also factors into that heavily. It's, "Oh, we're going to save 20 cents by running our container over here for $2,000.00 in data transfer to get the four terabytes it needs over to it." The model tends to break somewhat unfortunately.
Corey Sanders: Yes, you have to be intelligent about it. You can't just be blindly multi-cloud; you've got to be intelligent about how to handle it.
Corey Quinn: Semi-relatedly, I would consider multi-cloud to encompass on-premise as well to an extent, and there are a lot of migration stories there. To that end, I wanted to ask you a little bit about Azure Stack. For those who don't know–and please correct me if I'm wrong on any of this–it effectively lets you run an Azure-style platform on your own hardware. I'm sure I'll get hate mail for this but an open stack that isn't terrible and you're pretty close. It's the idea of being able to take, effectively, the same primitives, the same API calls and have things running in your own environment as well as an Azure. I've seen a lot of interest in it. Are you seeing a lot of adoption?
Corey Sanders: Both. We are actually seeing both, and it's definitely not terrible. I will stray away from other comments about the Stack. You buy the full Stack so you buy the hardware with the layer on top, with the software on top. We have seen a lot of excitement and interest but in very specific usage cases. Actually, Edge has become a very interesting usage case where you do need the full cloud model, the full computation model that's offered in public cloud but you wanted proximity or in a low-connected environment, like the example on stage in the oil rigs being able to do a fair amount of analytics while still being able to run in that isolated environment.
We've seen a lot of excitement and energy there and then, certainly, regulated environments or regions where we're not deployed with regulated requirements. We see quite a bit of interest and excitement there, and the key value is that consistency. You've got that same application model that exists in both, and this is the same reason why we're excited. I'm excited about the IOT-Edge as well. It's similar sort of consistency model. Where there, it's this container-based application model, with Azure Stack, it's the full API and portal model, but similar sort of consistency that, I think, really, developers are looking for the right ones to deploy in both public cloud and intelligent action.
Corey Quinn: One of the questions I have around the idea of Azure Stack is, on the one hand, it's easy to hand-wave over it and imagine, "Oh, it's just like having an effective Azure datacenter inside of my own datacenter, and this is awesome." To put it very directly, there's a definite skills gap between the caliber of engineer who spends all day, every day running an Azure datacenter and someone who does this across three racks in a colo somewhere. Having been that person running three racks, I assure you I was not Azure-caliber assisted then to run these things.
To some extent, in Azure, if there is a hardware failure, things automatically will migrate. There's a lot of those Edge-case failures that aren't visible to me. In my datacenter, the exact opposite is true and I have to care about that. Do people dipping their toes into the waters of working with Azure Stack sometimes mistakenly blame Azure or Azure Stack for, effectively, their poor datacenter management practices?
Corey Sanders: We haven't seen it that much, but I'll explain why. It's because important. There's a lot of aspects of Azure Stack that are deliberately controlled, and controlled in such a way to make sure that it's sort of the random ways in which people run their own datacenter, which is fine for their own datacenters but it doesn't create a limited experience when they're deploying in Azure Stack because their expectation with Azure Stack is it looks, and feels, and touches just like public Azure which means there are a set of expectations and requirements and things we don't know.
A great example: You can't just run it on your own hardware. One of the things that we learned very early in the early days of Azure Stack was if it ran on its own hardware, you have all kinds of random failures, and random issues, and so on and so forth and you get into a lot of difficulties with customers' expectations surrounding that. This is where we came in and said, "No. Look, we're going to have to force this and it's going to upset some people because they want to go take it as software," but we're saying, "To get the experience that you're looking for, this is the way you need to do it."
Some of those expectations, those regulations, while frustrating to some customers, enable us to get that consistent experience. I would argue that is what has been the challenge for both us and other products for private clouds, has been the freedom and flexibility actually comes with a huge tax. There's a set of customers that will get them right there, super sharp, super strong and they'll have a lot of developers working on it, but to just pick it up and run it and then also have flexibility becomes sort of a shoot-yourself-in-the-foot experience. They're spot on that that is a challenge and something that we are very careful about enabling while also making sure that the quality of services is spot-on for the Azure Stack.
Corey Quinn: Durability and reliability is always going to be a hard problem. I don't know that you can ever get away from that entirely. A question for you is, as you see people discussing Azure in the general sense–throughout the community, throughout the industry–what do you wish people understood about Azure that they seem to not be grasping the way that you do?
Corey Sanders: From a durability and a reliability experience?
Corey Quinn: Across the board.
Corey Sanders: That's a broad question. Let me start with durability and liability as a kick-off. I think one of the things that, going from the beginning early days to now that has been, I think, most exciting to me and especially in the infrastructure and compute side, is our ability to take the amount of data we have with operating our own system and actually use AI for our own internal purposes. This is something that's really been sort of an exciting–and not something we necessarily sell as a product–but it's something that's an experience that we enable.
A great example of this: We have a huge amount of data coming in around hardware failure, and being able to predict hardware failure and the key aspect of having so much hardware and so much data about that hardware and being able to say, "This is showing signs that we've seen before," and, given that, we need to go make sure we move customers off this before they see the end, before they see that machine die.
When I think about the benefits of AI, for me, it's very close to heart because I can build a better service because I'm able to use all this data that I've got and take advantage of it. The same things apply for our security-based services, being able to take the huge amounts of data we've got about security models and security solutions and apply it. The reason why I think this is so interesting to me is it's entirely why what we're able to do is different from what customers can do by their own. It's why we offer this service that I think is so compelling because we have this amount of data. We have this amount of AI-mineable –
Corey Quinn: – scales and incredible asset.
Corey Sanders: Yes. Again, in the early days, we didn't really have scales and so we didn't get it, but at this point, being able to apply that scale to create a better service is really, really exciting. We've talked it earlier, the Zerochan–being at the infrastructure, being at the bottom of the stack, there's always this question of, "What's going to make it better?" and that, be security, better reliability, better availability, all of that, being built with AI models using the huge amount of data we've got, I think that's a super unique and interesting opportunity for us.
Corey Quinn: I would agree. I think this is the sort of thing that takes time to emerge as a use of these things. Whenever a new technology comes out, it seems that the first applications of it are generally ridiculous. Take a look at the internet itself. When that first became somewhat consumer-mainstream, it felt like the only thing on the internet was Star Trek trivia and adult content. People were struggling for a long time to figure out what that –
What's the old line? There's this trick I learned once. It was called buffering. There's a definite story of ridiculous Edge-case toys demonstrating the problem that then start to inform business decisions, the idea of having AI working on hardware failure rates, on looking for patterns, finding signal and noise. A common complaint about AI is that many shops don't have enough data to do anything particularly innovative or groundbreaking with that. To some extent, scale becomes a necessary prerequisite for some applications of this technology.
Corey Sanders: Absolutely, and I think you've also hit very interesting points about the progression of technology innovation and the growth of that innovation. I think even if we look into our own progression on using intelligence, and which is such a big focus for us today, and the scenarios that we cover today–the flying drone with the camera to go view a pipe–I think if I remember a year ago, I think we had an Edge device that could tell whether the picture was a cat or not.
I think the level of reality that's coming into our edge-based solution based on the innovation both on the AI side but then also on the ability to run these things on the device and the power that comes with it, it's changing. It's changing to being real-world things where there's not a lot of companies out there that are like, "I really wish I had a way to tell if something's a cat or not." That's not actually necessarily a real-world scenario, but the amount that they're saying, "I wish I could tell if a pipe was broken," is quite a few. I think we are getting to the point where these become real scenarios, and I think that's an exciting progression. Seeing that cat or not step into broken pipes, seeing that progression, that's sort of where we come in and can offer those services and, I think, really needs to grow.
Corey Quinn: Right, and let's say you do all of this in a lab somewhere and never publicize it, and you skip the cat step and you just have the, "Is this pipe broken or not?" people look at you and, first, do they burn you at the stake because what you're doing looks like magic–they have no idea how you got there–but it also means that you're developing things in-house at that time rather than letting the community weigh back in?
I would argue that, 10 or 20 years ago, Microsoft very well would have been the kind of company that knew all of this in-house. Today, I think it's blindingly apparent that you're not. You're embracing the community and heading in very interesting directions that even if, though some of us on the periphery, don't always agree with it, it at least is much more understandable, what's driving your decisions.
Corey Sanders: I think that the constant amount of innovation, an incremental innovation, versus this big drop of, "Here's the new world," instead, we're making incremental steps. We made announcements last week, made a bunch this week and we'll make some more in the coming weeks. I think this is not, "Here's everything and we've solved the world's problems." It is, "Look at the steps that we've taken. Look at how, in the areas where we're leading, look at the things that we're excited about." By the way, we'll be back and we'll have more for you, and give us feedback.
Corey Quinn: Wonderful. Thank you very much for joining me. This has been Screaming in the Cloud. I'm Corey Quinn. This has been Corey Sanders, and I'll talk to you next week.