Input sound file:
Session Name: Is This The Year For Azure Colored Clouds?
Announcer Om Malik Satya Nadella
All right, that was fascinating. So next up, we have a really cool fireside chat with Om and Satya Nadella from Microsoft, and they’re going to be talking about Azure, and exactly what’s going on as the Cloud ecosystem battles shake out, so please join me in welcoming them to the stage.
Om Malik 00:34
Thank you, Joe. How are you doing, Satya?
Satya Nadella 00:34
Great! Glad to be here.
Om Malik 00:36
What do you make of that talk by Andrew? Do you think there is room for all these low power, high volume processors in the data center?
Satya Nadella 00:47
I think his point was that high volume economics in the long run seems to trump a lot of things, and I think that’s a basic economic reality. For example, the PC revolution, if you think about the entire business that I run, which is the server business at Microsoft, benefited from the PC penetrating the data center. So I think that is going to be true, so what are the high volume operating systems, what are the high volume chipsets, what’s the high volume innovation in the data center at the rack level, and even at the data center level for sure, is going to have a bearing on how the war shakes out. So I think there’s a fundamental truth to his thesis.
Om Malik 01:30
I think there is a lot of conflict in that conversation, people say that some of the benefits of x86 on the data cannot be underscored, and the other side is talking about ARM as the wave of the future, so there’s always something to think about. Clearly you’re responsible for Microsoft’s cloud business, so you must be thinking about these things a lot. The cost of your infrastructure– do you have any ARM architecture servers?
Satya Nadella 02:03
As you know we have Windows devices on ARM, so we have a shared code base of Windows that runs across multiple chip architectures. We’ve had a history of having Windows run on multiple chip architectures, you remember Alpha and MIPS and we’ve had all of that. To me what’s most important right now is to really think about how do you bring compute, storage, and network, both at atomic level and at a data center scale, and use that resource very efficiently through the power of software. I believe obviously in distributed computing, but there is a cost in terms of when you start distributing things, there is a cost to be paid in terms of coordinating things. You have to be very careful about what’s the price/performance of a chip. Just because the chip is pretty cheap, and having lots of it, you are basically taking all that complexity and putting it in the software coordination. Sometimes it’s great, sometimes it’s not, so it depends on workloads and applications. So it’s a much more nuanced conversation. Then there’s price. Things are all static today, so the question is in a more dynamic, competitive world, what would it be? Time will tell. As far as being ready for it, absolutely we are ready for it.
Om Malik 03:26
Cool. Now let’s talk about this big scale you were just talking about, I was thinking that companies like Google have surged on just a massive scale of their operations, which allows them to rethink and imagine cloud at a very different scale. Amazon has its other e-commerce operations, which allows them to think about the world at a web scale. Do you think Bing helps you guys in any way? Even though I know it’s a very small–
Satya Nadella 04:00
It’s steadily growing, 17. 6% of share in the US, but overall, your point of first party applications– I believe in that in a big way. If you’re in the infrastructure business, the thing that you really need is apps. Quite frankly at Microsoft we have perhaps the most diverse set of first party web-scale properties around. You start with Bing, but beyond Bing, we have Xbox Live. There’s 45 million subscribers to Xbox Live, and that’s all driving our cloud. Office 365, which is one of the bigger communication assets out there, or SAS applications, Skype, Dynamics, which is more of a line of business application. So in fact if anything, I believe the entire architecture of your stack can’t get hijacked by just one app-scale app. Having that diversity, some stateful, some stateless, some more transactional heavy, some media heavy– having that diversity is what really teaches you the lessons you need to build a third party platform.
Om Malik 05:04
So what has Bing brought to Azure, for example?
Satya Nadella 05:07
I think if you look at search perhaps the best way to characterize the search business and search technology is it’s applied machine learning at scale. In other words, you’re collecting lots of data, you’re processing lots of data, to discern the knowledge from that data and then understanding queries. So it’s a big data platform. A lot of the machine management and cluster management infrastructure in Azure is straight out of the Bing codebase. We started out by effectively building the clusters and scaling the clusters, and managing storage compute networks using the same technology as Bing, and now the reverse is happening, which is Bing is leveraging all of the Azure technology.
Om Malik 05:49
It’s funny why Steve doesn’t talk about this, with the Wall Street people, that this is helping me lose money, a lot of it on internet services, but it helps our–
Satya Nadella 06:00
We talk about it, it’s selective hearing perhaps,
Satya Nadella 06:04
But it’s absolutely the truth. Bing first of all as an application is on all our devices, the technology, and it’s not even just the machine management, in fact the biggest place is everything that we are doing around big data processing. Even though we have Hadoop, the underlying infrastructure that we make resilient in our clusters is all part of Bing. There’s a lot of innovation that come out of it.
Om Malik 06:26
Let’s talk about your first-party application, you clearly feel that it gives you some kind of advantage. What kind of advantage does Microsoft get out of first party applications, in your case I’m guessing apart from Bing it would be Microsoft Office Live, is a big application.
Satya Nadella 06:47
Yeah, I think the notion of being a cloud provider without living the cloud lifestyle, which is you have to battle test it. You’ve got to be able to build something which is across 20 odd data centers with 200 other colocations and an edge network, and then living every day, every failure mode there is on the network, or on the compute or on the storage across all of this, means you’re real. Now, that battle tests our cloud infrastructure. One of the unique things that we do is not only do we provide that infrastructure as a service in Windows Azure, but we also take that same tech and make it available as our server product for anyone else to stand up their cloud. To me, one of the questions everyone asks is “How did you guys get so good at the hypervisor between releases?” The only reason we got better is that we were using the hypervisor on Azure, and when you have millions of virtual machines on the same hypervisor that you’re running on a daily basis, and fixing issues, improving performance, then guess what? Your server also benefits.
Om Malik 07:54
Right. So do you think you’re going to be able to offer Azure, which is outside of your cloud, and outside of [inaudible]? Will you be able to offer Azure services–?
Satya Nadella 08:08
Yeah, in fact one of the things that we’d started doing last year and it’s really picked up steam, when we announced the Windows Azure pack for Windows Server, it’s a mouthful, but other than that, the reality is that as a service provider, you can now stand up a cloud which is basically the same technology that we used to stand up Azure. More importantly it also allows you to have the provisioning framework which is exactly the same as Azure. Our vision has always been that distributed computing is not going to get locked up by one single North American company, or multiple North American companies in their own data centers. I think that’s too narrow a worldview to have. You need to be able to build infrastructure that you provide as a service, but you also need to enable others. You have to have consistency, just because things are distributed means nothing if it isn’t simple and easy for people to be able to move between those clouds, use the provisioning layer between these clouds in a consistent way, and we’ve thought of that.
Om Malik 09:13
One of the things I’ve always struggled with personally is when everybody talks about hybrids, you’re kind of serving two masters. One is the enterprise and very specific use case scenarios, and then there is the actual public cloud, which is app-scale. Private clouds really aren’t app-scale like Amazon is, or what you guys have built with Azure is. Isn’t it just simpler to focus on one thing, which is the public cloud, and–
Satya Nadella 09:43
The reality is today, let’s just do some numbers, just to ground ourselves. There’s 8 million servers or so sold a year. At the end of the day the clouds we were discussing are backstage. The clouds do run on servers, the shape of servers are changing, 8 million of them are sold. You count all the public cloud providers, probably less than a million get sold on a yearly basis to them, so the other 7 million are getting sold outside. Virtualization rate, which we say is all done, is only at 30%. So that’s the state of the world. Now you could say, isn’t there a secular trend for more efficient utilization, and isn’t the public cloud shared multi-tenant infrastructure the most efficient? No question and that’s one of the reasons we believe in building out this app-scale property. Do we think that every SLA for every workload in every regulated industry, in every geography will get served by, as I said, a couple of North American companies in however many data centers? I don’t think that worldview works, it’s too narrow. We can argue whether it’s 10/90, or 50/50, 70/30, who knows. But I do believe if you want to be an infrastructure provider, to say ” Just come to my data center, we’ll take care of you,” I think will work for a large majority, but not for everybody, and our goal is to be for everybody.
Om Malik 11:01
I’ll just make a little push-back on this, I mean if the public cloud is good enough for CIA, I think it’s good enough for a lot companies.
Satya Nadella 11:11
There seems to be a lot of fine print, I think I was reading–
Om Malik 11:12
But still, it’s the CIA.
Satya Nadella 11:14
Is it really, is it public cloud, private cloud, I don’t know what it is.
Om Malik 11:21
As far as I know, Amazon doesn’t have a private cloud offering, so I’m assuming that this is their public cloud infrastructure.
Satya Nadella 11:31
That’s a good point, that is exactly our strategy, which is it’s not our private cloud infrastructure is some forked version of our public one. In fact, take my database team. There is no codebase called SQL server. There is the SQL database codebase which is on Azure. We reverse integrate that codebase to produce a server product, so there’s one stream of innovation. That makes a ton of sense to me, and that’s what we absolutely believe in. Otherwise then there’s no such thing– if someone wants to stand up their own government cloud, or a regulated cloud in the financial industry or what have you, they don’t want something that’s not first class, that’s not the stuff you’re using yourself to run your app-scale operation. From a design point, I absolutely believe in it. Whether it’s something that only one corporation or a few corporations can run, that I think is too narrow, from my point of view.
Om Malik 12:28
Talking about competition, let’s talk about when the early days of internet infrastructure and data centers, there were four companies that people would talk about all the time. It was Sun, Oracle, Cisco and Microsoft. Then there was HP in there somewhere, but essentially those were the four horsemen of the internet. Who are the people you think are the four horsemen of the cloud era?
Satya Nadella 12:56
Om Malik 14:35
It’s funny you don’t bring up any of your ecosystem partners, Dell, HP, in this conversation, do you think Dell, HP, IBM, they’re just going to be roadkill in the cloud wars?
Satya Nadella 14:51
I don’t– I didn’t say that, those are your words, I think.
Om Malik 14:54
No, I’m just asking.
Satya Nadella 14:58
The thing is, HP Dell, and IBM and others, have built systems and no-one should take lightly the expertise and depth they have. There is still last time I checked real systems and real hardware underneath server software, underneath cloud. Do they have a role to play which is significant as this ecosystem evolves? Absolutely. Are they also trying to do other things, in terms of becoming their own software and cloud providers? Absolutely. But are they at scale today? No. The folks I mentioned are the folks who are really defining and driving across the various layers the software assets. Is that the only list? What we’ve learnt in this industry is never fall in love with any list. This business is all about not longevity, but relevance and constant innovation.
Om Malik 15:48
So two things you are saying, there is going to be an ecosystem disruption, you just don’t want to name names–
Om Malik 15:55
And second, without having a really killer app, or killer apps on your platform, you really don’t become a platform.
Satya Nadella 16:03
Absolutely. The last one, my words.
Om Malik 16:08
Just rephrasing, seriously, I’m thinking about the world we have today. You have companies like Quanta. Nobody’s heard of them, and there they are becoming a major player in the industry. It’s happening at the network level, it’s happening at the storage level, people who are essentially OEMs in the past are coming directly to people like you, that’s a big change in the industry. Without naming names, tell me what happens to the client-server and PC ecosystem going forward.
Satya Nadella 16:45
See, this entire notion of taking a redundant array of inexpensive stuff, and then controlling it through software to build out these complex data centers, is what we have focused on, and that’s what an operating system does. It essentially manages resources and exposes an application platform. We think that the disruption that is happening, or the innovation that is happening in terms of being able to take commodity parts and through the power of software deliver real high quality SLA, is the design point. Now, who are the folks who are participating in it, there are the new players like the ODMs, there are ODMs doing very disruptive work, so we buy from OEMs and ODMs in our own data centers, we design systems with them, so every one of the OEMs you’ve mentioned also have their own web-scale divisions, where they’re also innovating like an ODM.
Om Malik 17:47
So one of the things– I used to cover telecoms, so one of the things that happened with the post-2001 telecom meltdown was we saw consolidation of telecom companies. There became maybe four or five buyers of telecom equipment, and the impact of that was we had guys like Lucent and Nortel is gone, Lucent is on it’s back right now. People who are doing well are actually government subsidized companies like ZTE. You start to see that, and I see a lot of parallels in the cloud, too. If the four of you, let’s say you, Amazon, Google are the four major players of the cloud, there is very little room for the vendors in there, the hardware vendors.
Satya Nadella 18:39
At the end it sort of goes back to– of course, the simple economic theory is that if the buyer power is concentrated on any part, there are going to be changes. But then the notion is when is this 8 million servers that we talked about going to become 8 million servers to these four. I think that’s a long ten year horizon, how will it shake out– one of the things that we don’t talk about is what’s going to be on the edge, even if everything is powered in the public cloud, what is the need for things on the edge, and what shape will it take? So I think there’s a lot of innovation to be added.
Om Malik 19:17
So before we go, the question I want to ask you is– again, it’s related to the telecoms, the biggest challenge we have in the world today is not building those fancy data centers, it’s you and I as consumers connecting to the data or the cloud services. It’s controlled by monopolies, they want to work with the old way of doing things. Do you think that’s a huge challenge for the cloud industry itself?
Satya Nadella 19:49
Broadly I think, if I had to sort of say, “What’s going to make cloud-computing more ubiquitous?” I would sort of really say it’s really the apps, that’s the driver. There are all sorts of impediments, connectivity, connectivity between data centers, when people think about moving, especially in enterprise, going to move an application on to the cloud, they care a lot about if they have to rendezvous backwards business logic in the enterprise, what’s my connectivity, can I do it over the public internet, or do I need to have some other proprietary link and so on. But the real impediment I think to adoption of the cloud, will really turn out to be at the end of the day the apps that get built to take native advantage of the cloud. That then will take care– the market will take care of all these other impediments. Basically there will be economic interests for people to take care of those things. The big wild card is how do all the regulation and privacy laws around cloud computing settle? That’s one of the reasons why I believe that to prematurely declare one form of deployment of cloud as the winner would be wrong, because I think there are many interests here.
Om Malik 21:08
Great. I have to go, we’re over time, and I really forgot to ask about that shakeup in Bellevue. Maybe next time I’ll ask you when you’re back next year. Thank you.
Satya Nadella 21:21
Sounds great. Thanks a lot, Om.