At TechEd 2013 earlier this year, Petri IT Knowledgebase contributor Aidan Finn and I had the opportunity to sit down for an interview with Microsoft’s Jeffrey Snover, a Distinguished Engineer and the Lead Architect for the Windows Server Division.
Microsoft had just announced Windows Server 2012 R2 and System Center 2012 R2 at TechEd, and most of what we covered in our discussion dealt with the new features in these updated products, where the IT industry was heading, and the growing importance of PowerShell. (Be sure to check out our What’s New in Windows Server 2012 R2 and Windows Server 2012 Hyper-V R2 for more info on these updates.)
Jeff James: To get this started, could you give us an overview of the announcements made today [at TechEd 2013] and Microsoft’s vision for the future of the IT industry, and where Microsoft would like IT pros to focus their attention?
Jeffrey Snover: Sure. The things we announced today were, first Windows Server 2012 R2, System Center 2012 R2, and then the Windows Azure Pack. I’m sure we have some other ones, but those are the ones that I focused in on.
These technologies allow us to have an integrated platform that then can be deployed in these three clouds. The three clouds, the whole notion of it is that the different clouds have different attributes. A customer running a private cloud has infinite control. He can decide what hardware he wants to run it on, he can design the bandwidth, he can design the latency, he can put it in a room, he can put a lock on the room and he can control who has access to things. You have infinite control.
Then, if you want to, you can run [IT workloads] on Windows Azure. With Windows Azure you have infinite scalability and elasticity and say, hey, I’d like a server. No, I meant a thousand servers. No, I meant 200 here, and 200 there, and so you have more flexibility, low latency, scalability, and elasticity.
From TechEd 2013, from left: Petri Editorial Director Jeff James, Microsoft’s Jeffrey Snover, and Petri contributor Aidan Finn.
Then there are service providers. With service providers, the idea is that we’ll [offer them great customization options] with these new releases. You can go to a service provider, and maybe a customer isn’t that comfortable with the cloud yet, and so they can work with the service provider to get value added services where they’re going to bring the hardware, and help guide the customer to the cloud.
What differentiates us from our competitors is this three-cloud vision. A lot of our competitors have a take-it-or-leave-it approach to the cloud. Our approach is to say that we’re going to take customers from wherever they are and take them wherever they want to go. They want to keep things, and run things on the private cloud. We’re going to help them be successful with that.
If they want to outsource things to a service provider, we’re going to help them be successful with that. If they’d like to try running things in our data centers, we’ll help them with that. If they want to have a hybrid solution, and run some things here and some things there, we want to support that. Even if they say “I’ve started here and I’ve changed my mind” we want to support that as well.
That allows our customers to embrace this. There have been the early adopters of the cloud, you know, the geeks, you know, people just love the bits.
Jeff: Like web developers who adopted cloud computing early.
Jeffrey: Right. But the mass market, what they want is they want things on their terms, they want simplicity, they want things that just work. They don’t want a gazillion bells and whistles. They just want to say, “I’m here. I want it over there. You figure it out.”
That’s why I think one of the big differentiators of our approach to software-defined networking and virtualization is we’re all about trying. We go down that path to try and simplify scenarios, not make them complex. Not to say, “You’re not just a system administrator. You can handle all the complexity of network management, too. Put your seat belt on. This is going to be a ride. Free up your weekends, because you’ve got some reading to do.”
That is not our vision. Our vision is to say, “You’re going to work with your networking guys to get some base capabilities, and then from there we’re going to handle it for you and simplify things.”
Jeff: On the topic of cloud computing, it seems that some elements of the IT community have resisted moving to the cloud, with concerns from fear about losing their jobs or legitimate security issues or a host of other concerns. Where are we right now in terms of that spectrum?
Jeffrey: A couple years ago [Microsoft CTO of the Server and Tools Division] Dave Campbell addressed this, and he said, “These things come up, these major industry shifts come up, and people get them in focus. But in reality they take ten years to really bear fruit.” He said that a couple years ago, and he says, “I think we’re in year three.” We’re probably in year five. Dave’s a pretty smart guy, and that feels maybe right, I don’t know, plus or minus two years. These things do take a while.
What you see is Windows Server 2012 laid the basic foundations. In this release you hear lots about this feature and that feature, Hyper‑V vs. VMware, gorgeous stuff, Desired State Configuration, all these great features. But what I think is perhaps the most impactful to customers is stuff we’re never going to talk about, because they seem too low-level, which is to say, we’re actually taking the software and we’re deploying these things as private clouds. We’re running them ourselves.
As we run them, we’re like, hey, that doesn’t work quite right. Let’s fix that. That’s a crummy error message. How did I get in this situation? How do I diagnose this? It’s all that. How do we make it real? Literally, there are many hundreds of features, many thousands of changes, that go into things just working. Each one of them is so small. We’re not going to talk about it. This just works.
It kind of popped up in my session on transforming your data center. There’s one that was big enough, which is to say, now we have direct support for being able to go from raw servers to scale‑out file servers. You could always do that, and now it’s “What’s the big deal? I could always do that.”
The answer is “Yeah, but now you click a button here, and you click a button there, and you click a button there, and then it’s done.” It’s like, “Oh, what did you used to do?” The answer is, “Boy, roll up your sleeves, grab your rabbit’s foot, these things take a while.” There’s just lots and lots like that.
What you’ll see is that depending upon the scenario, there are some scenarios where it’s just a slam dunk. Unambiguous. The advantages are so compelling. You are in the cloud, no answers. There are others where it’s a gray zone, and in each release the bar changes. In more and more things, it makes sense for more and more people to take advantage of it.
Jeff: It sounds like you got quite a bit of feedback from your hosting partners, because it seems like there’s a lot of features and a lot of these features aren’t really being telegraphed. There are not really being shouted from the mountaintops but from a hosting point of view and therefore the customers’ point of view, these are pretty important changes.
Jeff: Maybe you can talk just a little bit about how the feedback you received from your customers that led to the biggest changes in R2, how that process worked?
Jeffrey: Sure. We’ve done a great job in going out and talking to customers prior to Windows Server 2012. We just did a massive amount of discussions, quantitative stuff, and we sort of talk about the year we put our pens down. That’s all true, but we didn’t stop talking to our customers after that. We kept talking to them, and to a number of them we gave early builds, and then we got measured feedback and we got feedback from them. A lot of people got real excited about what they saw, and the opportunity to actually run their business on this stuff.
If you take a look at the hosters, they are really simple beasts. They want to be able to buy inexpensive stuff, buy low, sell high. Buy inexpensive stuff, they want it to just work, they want customers to have great performance, and they want it to be easy to operate. Then they want to offer it to people in a way that’s easy for them to consume. That’s why you see a number of the things we did.
You run around hardware designs and around being able to simplify the scenarios, and being able to measure things, and the Windows Azure Pack…
Jeff: Which is a new name, since the product had a different and much longer name originally…
Aidan Finn: …Yes, Windows Azure Services for Windows Server.
Jeffrey: The great thing about that is that that was somebody’s job [to come up with that product name]. That’s what he came up with.
Aidan: Is he paid by the letter? [laughter]
Jeffrey: It’s like, “Thanks for contributing to the bonus pool.”
Jeffrey: Anyway, so that’s why we have a number of features in there, and what is great is that a number of those features, actually, the vast bulk of them helped with the private clouds as well. You’ll see like the Hyper‑V Replica is such a great example. One of the points I made in my “Transforming The Data Center” talk was when you talk about business continuity. I said, “Everybody here can afford to buy a couple of extra disks to get better business continuity of data.” But not many of us can afford an additional data center, or additional data centers in multiple locations throughout the world.
That’s one of the benefits of the technology that we’ve designed here. Hyper‑V Replica says, “Hey, I’ve just got a couple of servers here.” I can take advantage of a service provider, and that service provider can take advantage of another service provider or Azure to provide this guy this amazing capability. That’s one of the pretty exciting scenarios.
Aidan: It’s a great example of the whole one consistent platform in the three clouds. It’s also great for the [SMB] market as well, because here’s a solution for this small/medium enterprise who’s struggled with this business challenge. Now it’s bringing in the hosting company to the Hyper‑V world because they are looking at that opportunity… it brings the whole story together, just with that one feature.
Jeffrey: It does. The great thing about Microsoft is how good we at taking really high-end computing [features and capability] and making them available to the masses. If you’re familiar with Geoffrey A. Moore’s model, we cross the chasm. He has this model of early adopters, and then the mass market. He says there’s a chasm between them.
A lot of people were doing very well in the early adopters, failed to reach the mass market, and why is that? The answer is that the early adopters, they’re geeks. They love all the bells and the whistles, and all that stuff. These are the very same attributes that the mass market reject. They don’t want that, they want it to just work. I think you’ll see a lot of the players that are big in the cloud today they’ll just kind of fail away. Fade away, fail away. That was a mistake, but it turns out it was a really good one.
Jeff: Freudian slip there.
Jeffrey: They’ll just fail away.
Jeffrey: What Microsoft does, is we simplify high-end scenarios. We simplify them. We get them running on low cost commodity hardware that are high volume. This combination then drives adoption, which drives volume, which lowers price, which drives adoption.
This is the virtuous cycle that Microsoft just nails. We did it for the PC. We said we were going to do it for the enterprise. People laughed at us. Microsoft, the PC company, enterprise software. Enterprise software running on x86 servers, and now we’re dominant vendor in the enterprise software marketplace, and we’re going to do it again for the cloud. It’s our DNA. It’s what we do well. It’s what we do, it’s who we are.
We take these high end things and simplify, simplify, simplify. Don’t run it on the high end stuff. Let’s work with Intel, get it running on the inexpensive stuff, and go for volume. Do it again.
Jeff: SQL Server is a great example of that. Over the last decade, SQL Server market share has increased in larger enterprises pretty significantly.
Aidan: Even that’s gotten better as well when you start looking at SMB3 storage, so adopting those high‑end SQL applications running on InfiniBand networking. Got some pretty cool scenarios there.
Jeffrey: Yeah, it turns out that at the company we do these polls. Internal, employee polls, “Hey, what are we doing well? What are we not doing well? What do you like about your job, what do you hate about your job,” et cetera.
Then we act upon that. But year after year after year, for as long as I’ve been here, and I think everyone told me before I got there, the way at the bottom was, we don’t work together as a cross‑group cooperation. We’re not very good at that. Ever since we went, and got on this cloud OS vision. We’ve got this crystal clear vision for not a product team, but for the company.
The level of cross‑group collaboration is fantastic. Now, the poll is right up there at the top. We add the latest version, the latest milestone, we finished it. Jeff Woolsey held what’s called “Demo Days.” Which is he gets each one of the feature teams to go give a demo of what they did to all the engineers, we’ve got a big monster room. It was just stunning, one after the other, after the other. Get up, show this amazing feature, and then they said we would like to thank this team, and that team, and that team.
These guys provided this, these guys provided that, and these guys provided this to enable this. Next guy. We want to thank this team, and this team, and this team. The amount of collaboration is just phenomenal. SQL used to be its own operating system. Like, “Hey, Windows, get the heck out of the way, I just want to talk to the transistors. Anything between me and the transistors get out of the way.”
That’s completely transformed now, and why? The answer is we jointly worked with them on joint performance goals. Say, let’s work on joint performance goals. We’ve got a fantastic windows performance team, who can analyze things like you cannot imagine, rocket science level. They’d get there and say, “Okay, you need to change this code. You need to change that code.” Holy shmoley! Performance goes through the roof.
Now they’re a very good example of taking advantage of the operating system in a way that gives them fantastic performance. All the NUMA stuff, the NUMA scaling stuff with them really just worked well.
Aidan: On that theme, I think one of the questions people are going to be asking after today, they heard Windows Server 2012 R2, fantastic. System Center 2012 R2, fantastic. Will they be released together? Will it be one release? Will it be Windows Server and six months later, System Center?
Jeffrey: They’re going to be together.
Aidan: Cool. I think a lot of people will be delighted with that.
Jeff: On a different topic, I know you’re a big advocate of PowerShell. I know a lot of [Petri IT Knowledgebase] authors love PowerShell. It’s a very quick and easy way to automate various things and do things quickly. What would you say to any holdout administrators who don’t like to touch the command line? What advice would you give them? Or maybe you have some tips you could give them to help them down the path towards PowerShell enlightenment?
Jeffrey: I’d say the industry advances and people get to decide whether they want to advance with it or not. In general, most people in the IT field like that, that’s why we joined, right? One of my best friends, I love him, he’s works with lumber. You know what? He learned his business, but trees don’t change. He spends his weekend playing with his kids.
He’s got a great life, it’s wonderful. I chose a different path. I chose a path where I have to constantly work on my skills. We all have, people in the tech field, that’s what we do. We in general, the folks do that. Now that said, they don’t want to squander. They don’t want to keep chasing things unless they’re there. I think a number of people said, “Hey is this going to be a flash in the pan?” Sometimes these things show up, and then they go away.
I don’t want to invest and do that. We treat people’s investment in PowerShell with great respect. That’s why you saw when we went and did workflow, instead of having a new syntax, etc. We integrated it as part of the language.
I don’t know if it popped in what we discussed here, but now the Orchestrator team has taken our engine, and used our extension points and it’s the very same PowerShell engine with the extension points. That’s how they’re delivering the PowerShell as a service as part of the Windows Azure pack. I’d say get with the program.
Jeff: Any other specific features in these new releases that you’re excited about?
Jeffrey: I think the thing we’ve not as good a job communicating is just the dramatic transformational nature of our storage advances. We saw this in the studies: number one component of IT budget costs? Storage. Number one growing, storage. Happiness quotient? It varies. Some people are pretty happy. Those people spend a lot of money. They’re pretty happy. Others, just not happy at all.
If you take a look at our storage stack, I think, prior to 2012, we did not have a strong story and so I think when we came out there, people weren’t even listening, perhaps, because we were just on a spot of weakness. We did not have a strong story. But 2012, we had a very strong story. In 2012 R2, we throw gasoline on that fire. There’s no ambiguity. From an engineering and architect standpoint, the ability to manipulate data at massive bandwidth, with low latency, with fantastic eye‑ops is the key to computing.
We have invested accordingly, and it has paid off. I don’t know if you saw my slide, the guys had originally did it. They said, “Great price/performance trade‑off.” I said, “No! You don’t get it! That’s exactly incorrect!” We reject price/performance trade‑off. We’re giving you great performance at fantastic price. There’s no trade‑off involved here. I’m taking cheap, cheap, cheap stuff, so great performance and reliability, robustness.
The example I gave was IOPS or rebuilds. Let’s say you had mirroring. Here’s the observation, the observation is that IOs don’t change, IOs per second, but capacity does. If I’m mirroring, and I’ve got a one terabyte drive and this dies, I have to use the second drive to create the third drive, and it’s one terabyte. The IOPS don’t change. However one terabyte divided by the number of IOPS, that’s how many reads I have to do, that’s how many writes I have to do. That tells you how long it’s going to take, because the ops don’t change. They haven’t changed forever. That’s how long it takes.
As I go to two terabytes, now I have twice as many IOPS that I have to perform, or IOs I have to perform, which means my time doubles. If I go to four and then eight terabytes, that window of exposure keeps increasing. Now, two problems. One is I was actually using that disk, so the fact that you’re beating that, consuming all the available IOPS to do a rebuild, I’ve got a problem with that. Two is wait a second, what happens if this disk has a problem? You’re out of luck. How long can it might have a problem? The answer is, as the disk grows, that window grows very, very large. That’s a problem.
What we do is we do striped. We have very large striped sets, 100+ disks that add right to a file. When a disk goes down, what I can do is in this release, we have parallel rebuilds that says, “I’m going to get the blocks from a hundred other disks and rebuild things, spread that out.” I’ve got lots of these disks contributing.
What that means is twofold. One, when something goes down – I don’t know if you’ve experienced this in other systems – everything’s fine. I joke, things work when they work, and fail when they fail. Everything’s working, but then when something fails, other things start to fail and other things start to fail and other things start to fail.
In this event, for reasons like that, in this environment, something fails, all of a sudden I’ve got a large number of resources all contributing a small amount to the resolution, and it gets resolved very, very quickly. There’s just tons of things like that. Anyway, I think our IO story, our storage story, is the big thing. Not expensive storage, we’re talking cheap, cheap, cheap storage.
We did the work with the SQL guys and the performance guys. They’re awesome, they’re nerds, they’re just, like, total nerds, they have no presentation skills whatsoever. We’re sitting there and going through, blah, blah, blah, blah, blah. I’m, like, “yeah, yeah, yeah.” I look at this and I’m, like, “I don’t think I understand what you’re telling me.” I said to him, “Do I understand this correct? Because this data seems to indicate that SQL running against local disks is, in some cases, slower than SQL going against an SMB file share?” They said, “Yes.” How is that possible? A bit of a sampling, and you got multiple CPUs. I said, “Listen, what you’re saying is that accessing a remote database is as fast as a local one?” They said, “Yes.”
I said, “The only appropriate reaction to that is to be jumping up and down on the tables. That’s crazy! Why?” The answer is we fixed our storage stack. We worked with these guys and integrated. That’s just a phenomenal achievement using inexpensive stuff.
Jeff: There’s a lot of storage vendors in the market, like EMC and that app, and all these companies that have their own proprietary systems and they have their own way of managing storage sometimes? How does that impact them?
Jeffrey: What we did in 2012 is we embraced them. Prior to 2012, if you had one of these systems, how did you integrate it into Windows? My joke is with a rabbit’s foot and a bottle of aspirin. It’s really kind of hard, you had to get the software, get all these systems. We now support those. If you decided to do that, we support them. We invest heavily to make it integrate in easily. It’s actually my first example of this data center plug‑and‑play.
We have an SMIS service, you install a storage area network, plug it into your environment. We will find it, we will enumerate its resources, we make those resources available in the same way we do our local resources, through the same set of APIs and power shut commandlets, inbox GUIs and system center tools. They have full support. Light up their stuff.
We also added extension points to highlight their capabilities, things like offload data transfer. We’ll continue to do those things, but as our storage stack gets more and more capacious, leveraging in more inexpensive stuff, clearly it’ll be a challenge for them, and they’ll have to be clearer about their value proposition.
Jeff: To add more value, basically.
Aidan: This isn’t throwing away, like you said ODX trim on map, and also surface those features through your SMB3 file shares, so you can leverage that block storage and your SMB3 scale out file servers.
Jeffrey: I’m sure they didn’t talk about this, our approach versus the competitors. We view this as a cloud operating system problem. The cloud is an operating system problem. What that means is we’ll make any change at any layer to solve the problem correctly. Our competitors use the cloud as something you built on top of an operating system. Therefore they can’t do those things. What we did was we had RDMA, we integrated it into SMB, and having done that work, we’re in position to say, “Hey, let’s get live migration over SMB over RDMA.” It becomes very easy for us to do.
Now all of a sudden, crazy leverage ability to live migrate an incredible amount of stuff with extremely low latency and low CPU usage.
Jeff: While things are running.
Jeffrey: Yeah, it is.
Aidan: [It also will help] reduce the costs of that data center and that hosting company as well. The the cost of RDMA and networking is going down. You’re using fewer switches, less electricity, so that the cost of deploying your host and revealing your services to your client is that much lower.
Jeffrey: That was one of the examples that I just love, because it highlighted our architectural investments paying off. It turns out that wasn’t a big team doing that, it was small. Why? They were able to deliver a huge feature, because of the architectural work we had done for previous reversions.
Aidan and I would like to thank Jeffrey for taking the time to sit down for the interview with us, and we have more Windows Server 2012 R2 interview content coming up shortly. Look for our interview with Microsoft storage guru Elden Christensen, where we’ll talk more about some of the new storage features in Windows Server 2012 R2. We’ll be posting that in the next few days, so keep watching this space for more updates.