The Last Snowflake - 2011 Uptime Symposium

George Slessman, CEO of IO, gives a presentation entitled, 'The Last Snowflake' at the 2011 Uptime Symposium

Transcript



Moderator:

Thank you, Bruce. We have the opportunity at this point to kind of do a quick test if you will. Has anybody here heard or internalized a sense that processes are changing in how we deliver data center space or are going to change shortly? If you haven't heard that message, we have another opportunity for you to hear it now. I think this is a very valuable piece of insight that you're going to hear in this next segment, and no the title does not suggest that George is our next metrological consultant with the last of the data centers snowflakes.

George Slessman is an industrial engineer and an entrepreneur, now CEO of IO Data Centers, that is one of our Diamond sponsors, and I am delighted to have the opportunity to introduce George, because 18 months ago I think we thought one another were the enemies, and then we find we have the same views, the same urgency about things that need to change within our industry to keep it healthy, vibrant, and affordable. With that George.

George Slessman:

Hello everybody, I'm excited to be here, so excited I'm going to leave my jacket off but I want to bring it just to prove that I have one. Actually I'm humbled and tremendously excited to be here today. I look across the crowd here and I see a lot of people that I have a tremendous amount of respect for and respect for what they've done in this industry and has led up to where we are and what we're looking to accomplish here.

The Uptime Institute and to Pitt's comment, I think for many years a lot of folks in the colocation and wholesale data centers space, from an outsource provider, had felt that the Uptime Institute and our interests hadn't aligned perfectly, and I think that they actually do align perfectly because customers that we focus on, emphasize the enterprise users, have been the critical participants in what the Uptime Institute does and what the mission is, which is to keep uptime.

At the end of the day, it's in the statement. We're looking to keep applications alive by whatever means possible, and right now and the way we've accomplished that in the past has been what I'm going to describe here as snowflakes, not to, as Pitt said, speak of the meteorological outcome of snowflakes but simply the outcome of what a data center has been.

Additionally, I see some other folks. I see a bunch of my team members here. I couldn't have accomplished anything that I've accomplished or we've accomplished as an organization without the various people that are sitting here from our engineering team to the folks on our sales team and mentors. I see Peter Gross sitting here, who everyone I'm sure knows, and has been a recent person I consider a friend these days. So I really appreciate again all your time and effort.

So feel free to blurt out anytime if you think I'm full of it, if you don't think I know what I am talking about. I enjoy the dialogue. I'd love to hear your feedback.

So the name of the presentation is "The Last Snowflake," and I guess at first I want to talk about is what do I mean. So a snowflake, and I'm showing a data center under construction here. The concept is and the reason I'm mentioning this and what I mean when I'm talking about snowflakes is to solve this data center requirement that we've had, this application requirement to keep applications alive, we've essentially been building custom one-off data centers, one at a time, essentially snowflakes. Every single one of them looks different, every single one of them acts different, every single one of them is operated differently, and now purposefully they've operated very successfully and they've achieved the goal over time, I would say, relatively effectively of delivering on data center uptime or application uptime, keeping our friends and foes, the users with their irrational expectations for service delivery happy to date. The point of this is what our view is, and my view, and what I'm going to talk about today, is that I believe the last facilities based data center "snowflake" will be constructed in the next 24 months. Essentially, we are at the transformative stage of seeing the data center become part of the technology stack. We are at a point in time where essentially not unlike, let us call it the year or five years before EMC came into existence, when storage was managed with devices stuck inside of servers, to where storage became such a transformative component of the IT stack that it had to be managed in freestanding, purpose filled devices managed by software in a standard platform, or when we saw the utilization of the hardware stack go to a point where it was no longer manageable, so VMware came into creation to essentially optimize and drive utilization across the hardware stack.

It's my assertion, my belief, that we are at the precipice of the same transformation in the data center community. How we think about and deliver data centers is going to change to the point where essentially facilities based data centers are not going to be built again.

Digitization. You ask why, why do I believe this? The reason this matters is we start at the very beginning. The First Industrial Revolution, maybe you think I'm getting a little carried away and I'm talking about things that don't make a lot of sense, but in the First Industrial Revolution what we did was we mechanized every single economic process on planet Earth. From communications, we mechanized agriculture, we mechanized the food services industry, we mechanized transportation, we mechanized all of these things that we do and we did every day, and we still do today, we mechanized.

In the process of mechanization, we created industries, production manufacturing. We created massive energy consumption. Energy concentrations changed. We went from a cycle time . . . pick a transaction event, pre- mechanical First Industrial Revolution, farming, it was a person with a plow and a horse or an ox, that was plowing the field. Their ability to plow and manage this was essentially the cycle time was X. Post mechanical revolution, that cycle time was cut down in quanta. We are talking about quantum shifts in the way things are done through these revolutionary cycles. Essentially, you went from one person being able to farm an acre to one person being able to farm hundreds, if not thousands of acres.

We have seen this quantum revolution occur in the past. Another industry that was driven out of this entire outcome was the energy industry. The oil industry was created to support the mechanical revolution, because energy densities had to reach a point where you could support these types of transaction events. That essentially the ability to plow a field required complex hydrocarbons, because there was enough energy stored in that when you exploded it in a combustion engine, you could drive these types of processes.

So we get to the current industrial revolution. I'm going to go ahead and skip to the next slide. Digitization, we are taking every economic process on planet Earth, and we are digitizing it. We are taking it from a mechanical cycle to a digital cycle. We are taking these same transactions that used to exist and take minutes, which at one point in time took days if not years, and now we are doing them in seconds, if not milliseconds, if not microseconds, then nanoseconds.

The point of it is that we are massively transforming the way we do business and the way we conduct these economic processes. An example I use is banking. Let's think about banking. Literally, banking was the epitome of "I don't work a lot" if you worked in banking, right? "Banking hours" was from 9:00 in the morning to 3:30 in the afternoon, Monday through Friday except for bank holidays.

Now, think about where we are today. You can literally take a picture of a check, sitting in your hotel room in Singapore, and deposit it in your bank. Banks now are conducting more transactions electronically than they process manually in branches. We went from a time where you could conduct a transaction only for a few hours per day, at a physical branch, to being able to transact it 24 hours a day, 7 days a week, 365 days a year, whenever and however you want to.

The point of this graph is this is the beginning of revolution - 1969. We can even take a little bit of a step back from there. I'm going to skip into the math here, real quick. Let's go back to the ENIAC, 1944. One computer on planet Earth, one application. Beginning of the revolution. How many devices are there today connected to the Internet? 50 billion? Some enormous number?

If you go back to the network node slide, there are 900 million nodes now connected to the network. Look at the cycle times and look how quickly this is escalating. They don't even take the time to show that were four hosts in 1969, on the Internet, and now we're at 900 million.

The underlying thesis of this, and the only reason this matters, Digital Universe, a couple more facts. This is why we talk about this. This is from IDC Digital Universe Study, sponsored by EMC. Obviously, it benefits EMC to think there are going to be 35 zettabytes of data on planet Earth. The point is that, even if it's off by a factor or a couple factors, it's still an inordinate amount of data that we're going to be processing and managing on a daily basis. From 0.8 zettabytes today to 35 by 2020, that's 44 times the growth.

If I go back to this slide here, the doubling time is every one and a half years. Every 18 months the amount of computation that a computer can do per kilowatt hour of power is doubling, and we're still installing more servers. If server growth is growing at 8% to 10% per year and we're doubling the computational power of those servers and we're driving utilization through virtualization, what this means is there's the most precipitous gain in transactional volume that we've ever seen in the history of the world.

Where the rubber hits the road, then, is in energy consumption in the back of the server, because these transactions cannot be conducted without charging the capacitive logic gates inside of those microprocessors. We have to put energy in these.

The other dirty secret, I wouldn't say dirty secret. I don't know if it's even a secret. Obviously, if I know it, it's not a secret. Energy per instruction is not getting better. Energy efficiency of chips is getting better, just like cars are getting more efficient. But, if the car's driving more distance, it's consuming more aggregate energy. So what if a car goes from 18 miles a gallon to 50 miles a gallon? If you're driving 25,000 more miles this year, you're going to consume more energy.

At the most fundamental level of the chip, the chip, over time, is not consuming less energy per instruction, as scalar performance increases. If you want your computer to go faster, you want the cycle times, literally cycle times. I described cycle times as the times it takes to plow a field to the cycle, the clock time on a processing chip. If you want to do it faster, it requires more energy, and not linearly. From the i486 chip to the Pentium 4 chip, eight times the scalar performance, 37 times the energy per instruction.

You layer all of these things together. This is, again, from Intel's Director of Tera-Scale Computing. The world generates more data than it can store. That's his assertion. Data creation is rising 60% per year. By 2012, data created will be 2X what can be stored at any given moment. Walmart adds a billion rows per minute to its database. Per minute. Then all of the ancillary processing power to manage it, so if this is Walmart today, this is every corporation in the world five, ten, fifteen years from now.

Metadata is going to be coming from everywhere. It is going to be created everywhere and managed everywhere. So, medical imaging databases can acquire a petabyte. An individual human genome, according to a friend of mine who founded Braxton Health Care, is two petabytes of data per human genome. When you sequence an entire genome, it is an astronomical amount of data per genome. His objective is to sequence everyone on planet Earth. Where is all that data going? Who is going to run all of the processing power?

The Hadron Collider can generate a terabyte of data per second. In 2009, despite the recession, digital storage grew by 62% to 800,000 petabytes, which would be the equivalent of a stack of DVDs reaching the moon and back. By 2020, the stack will reach half way to Mars.

Then here is the punch line, and I apologize for it not being terribly easy to read. This is just some silly, back-of-the-envelope math that I did in Excel, just to prove the point. So, if we continue to build data centers the way we build them today, one megawatt of usable IT, Tier 3 enterprise data center, design specification, and we run it out for 40 years, and we assume that annual IT growth is 20% per year, a relatively rational number, it has been relatively consistent over the last 20 years, whether you look at Cisco, Intel or any of these folks on the generic IT spend, which is indicative of what is going into the data center. You then assume what has never happened before in the history of the world, that aggregate construction costs go down 8% per year, every year for the next 40 years. It has never happened in the history of tracking construction costs, construction costs have never gone down over an extended period of time, which makes sense, with inflation. This is not inflation-adjusted. This is just assuming core numbers. You assume that one megawatt costs $15 million per megawatt. Total cost of ownership to construct, I am not even talking operational expenses, I am just simply talking about construction. Take those numbers, the purple curve is the annual dollars spent to keep your data center capacity in tow with the energy consumption that it is going to require. The green line, shrinking down asymptotically to zero, is construction costs.

Even with construction costs, essentially at the end of the day reaching almost $800 a megawatt, or $800,000 a megawatt is the number it ends up being at the end of this, you are still spending hundreds of millions of dollars a year on a 20% compounding scale, 40 years from now, to keep the data center operational. And forget about going out 40 years. Let's just go out a couple of years. Let's go out three years. Let us go out four years.

So this is the point of my view, and our view as an organization and the team at IO and what drives us every day for what we are doing is we looked at this problem a year and a half ago, as we own and operate huge data centers, or big data centers - a lot of you operate much larger data centers than we do - and said, "How do I not end up owning a collection of 8-track tapes?" I could have the biggest, meanest collection of 8-track tapes on planet Earth five years from now, and no one is going to pay me a dime for them. No one is going to care. Someone is going to invent a technology to solve this problem, and I'm going to be standing there holding the bag. I said, "So let's be the people that invent the technology to figure this out. Or let's at least present a solution. Let's find a way to solve this problem today."

The punch line: Snowflakes are too expensive and take too long to deliver to meet the demands of this revolution. That if we play out this revolution, if we believe that imagination is all that dictates our ability to create, that 14 months ago this device didn't exist, how many of them are in this room? Everyone have one? I've got two, so I count for two people, right? Because I can't give up my last one because I haven't synced it with this one yet.

So the point of it is that devices outnumber people. We haven't even touched the developing world yet - India and China and Brazil and Russia. These countries are going to be adopting these technologies as well and using them the same way we are.

So the solution, the solution starts with what the objective is, right? So taking a classic, what I would hope to be an engineering approach to this, why does this matter? What are we trying to solve? Where are we going?

The data center is IT. We've called all sorts of things. We've declared it things it isn't. We've called them buildings. We've called them real estate. We've called them . . . who knows what we've called them. We've called them modular. We've called them containers. We've called them all this stuff. At the end of the day, the data center supports IT. Without IT, you wouldn't have a data center, and without users, you wouldn't have applications. Without applications, you wouldn't have hardware. If you didn't have hardware, hardware wouldn't be consuming energy, and that energy wouldn't have to be on all the time. Again, back to if we didn't have irrational users, right?

So the point of it is that we have to look at the solution. An objective of a data center is to keep the application and the users connected all the time. So IT always has to be on. So if we take these two propositions that we're now dealing with. Data centers are dis-economic as they're constructed today, and it's not saying we didn't make good decisions. I'm very proud of the data centers that we've constructed. I'm very proud of everything we've done from a technology perspective and how we operate, and everyone in this room should be as well. Just things are changing.

This is the time to adopt new technology and see where everything's going. I loved the day too when you have one server with one application on it and you never had to worry about it. You turned it on, both power supplies plugged in, you plugged in the fiber channel, you plugged in this and you're done. No one ever stepped on your resources or ran over the pool on the storage array and crashed it. That was great, but at the end of the day it didn't meet the objectives of the IT department was that we have to create a sustainable IT environment from an economic perspective as well as an environmental perspective.

So you take the fact that these costs are not going down, let's just assume they stay flat, it's unsustainable, and we have to keep the application alive all the time. We come to a solution.

First thing we have to solve for is standardized hardware. You can't possibly begin to solve this long-term solution if you're trying to solve it for snowflakes. We want to build ice cubes, standardized hardware. We want a thing you stick in the fridge, you crack it, break it open and out falls 8 cubes or 16 cubes or 22 cubes, or whatever the number is, in base 8 of course. The point of it is, is that you need standardized hardware.

So we endeavor to create standardized hardware for the data center. When I say standardized hardware, I'm not talking about the IT stack. I'm talking about the physical data center infrastructure. So the box, the enclosure, the container, the module, whatever you want to call it, the thing the IT stuff goes inside of, we have to build that in a standardized format. We have to take all of the infrastructure necessary to support that application and keep it on all the time, and we have to build it into a standardized set of components with a standardized set of discreet logic controllers so that now we can start to look further down the road and continue to solve this problem, which is software optimized and managed.

The next step to the data center, once you have created this standardized hardware platform, to really push the costs down, because look, you take a data center and break it up into pieces. You take a data center and just spread it out on the floor. This is a classic engineering treatment. You take a car apart. You break it into all of its pieces, lay all of the pieces on the floor, and then you realize, when you put it together, there's still a part laying on the floor, and you don't know where it goes.

The point of it here is, to get the price of data centers down to where it needs to be to be economic for companies to afford it, we have to eliminate components completely. Static UPS's, gone. Prime generation for every application, gone. I'm not talking next year, I'm not talking five years. I'm talking about as we get to that curve, as you see it moving out, to make the data center or make the energy delivery cycle economic for customers to utilize it and buy more servers and buy more network devices or buy more storage devices and keep up with what's happening, you've got to eliminate some of these components.

How that happens is software. This is going to be an application-managed environment. The software I'm talking about here is what we call our operating system, that takes this standardized hardware platform and layers a piece of software on top of it that manages generator, chiller, power conditioning, power delivery, energy recovery, efficiency, systems, free cooling.

All of these attributes that you want in a data center, now, we can manage from an automation perspective. That's interesting. We've done that before. It works just fine. Hell, you can take a building management system and make it do that.

The next step is making the data center managed. And when I say "managed," it's making the IT stack cognizant of where it lives, and making where it lives cognizant of what lives in it.

Right now the data center has no idea. The reason that you can be an effective colocation provider today - that's the world I grew up in - is because I don't care what happens in the IT stack. I can just simply be deaf, dumb, and stupid, building some stuff, manage some generators, keep fuel oil in it. You can show up with your IT stack and you can be deaf, dumb, and mute with regard to what the data center does.

This plays out in a colocation environment, as well as in your internal environment. How many of your software developers really know what runs the data center? None of them, by definition. How many of your data center guys, that make sure the generators, chillers, and UPS's stay online, know how to write a line of code? No one does, but this has got to merge to make this happen over time.

We have to improve operational sustainability by making these things standardized - hardware standardization, software automation, linking that software into the application stack. VMware, now, can talk to the IOS. The IOS can say, "Hey, module 17 is in a yellow state. vMotion your applications to module 12." This is the type of interaction we're talking about.

Then, from an operational perspective, no matter how great your technology is and how fascinating, you cannot deliver on a service level without great operational sustainability. You have to have the people making this happen. No matter how great you engineer an airplane, how many fail-safe redundant systems you find, if a guy flies it into the ground, it crashes.

One of the benefits of the system and of this standardized platform is you now create a standardized platform operational sustainability. Your data centers look the same in every single market, whether you're in Singapore or you're in Poughkeepsie or you're in Dallas. You blindfold your guy and ask him where the air handlers are, and he knows where they are. He knows how to fix them. He knows what the playbook is, because the playbook's identical. If he phones home to find out what he should do or how he should it or she should do it, the answer is there. At the end of the day, what we've deemed this and what we call this and where we believe the transformation leading here is what's called digital energy technology.

This is the technology hardware, software, integrated into IT stack, to keep all of these things that we want to do every day, whether it's as commercial consumers or consumers of IT services, Facebook, all of the tools that we use at home, email, just our home email, sending pictures off to folks, all of the things that we do at home to all of the things we do commercially, the underpinnings of this that are going to keep this alive is a sector of technology that we call internally and how we refer to it as digital energy technology.

The reason we changed the vernaculars, we have to think about this differently. We talk generators and we talk to UPS's and we talk all of those things every day. That's what we do, and we own a lot of them, a lot a lot of them. Dave Shaw, our COO, is sitting here in the front. He can vouch for how many UPS's he's responsible for keeping online, how many generators we have, and how many are sitting there empty.

The real problem with data centers today is fundamentally when you stand there and look at the screen telling you what the utilization is. Right? We've all done it. I mean you stood there. No one wants to talk about it, but you go into a beautiful 10 megawatt Tier 4 data center and it's running at 24% of utilization. Best it can do is 50%, right, to maintain it's sustainability from a resiliency perspective. We've got to solve this. We can't be running data centers. The largest user of data centers globally, the enterprise consumers of data centers, the four global 2,000, cannot be running their data centers at . . . everyone thought it was horrible when servers ran at 20% utilization.

Let's dig it down a little deeper. The outcome here, manufacturing process being wrapped in staging, packed with IT by the customers, placed on a truck, shipped to wherever it needs to go, wherever you need data center, cycle time from day it starts to day it ends, 90 days from day you order, an entire 3 megawatt data center can be set up. This is one module. A data center is 15 modules, in our speak.

Deployed in the site, what you get inside for your IT equipment, four foot cold aisle or three foot hot aisle, fully automated control panel built in, discrete automation controllers, fire suppression, all this stuff that we hopefully a few years from now no one's going to care about. Managed by software, energy recovery, power distribution, ITN, interface to the virtualization layer, interface into your ticketing system, management, discrete automation management all the way across the system, optimization, simulation. All of these things are going to come true here certainly.

I am going to be able to rip a copy of your data center operating environment, have it run in a simulation, pre-populate it with all of your new Intel and HP gear, and then you can go back to the HP, Intel, IBM, and Cisco and say, "Yeah, you said it was going to run more efficiently but it doesn't. And here's why." Or it does, and this is why I'm going to buy HP here and this is why I'm going to buy this IBM here and this is why I'm going to do what I'm going to do.

The punch line here is just in time. This is all arriving just, and I say just in time in more than one way. Just in time means it's showing up just in time to solve this problem. We have 112 modules that are in production and committed to customers in the next three months. We launched the product in August of last year at Symposium, not Uptime Symposium, another Symposium which we shouldn't have in San Francisco here. Opened our factory in February, started marketing the product, and 112 of those things you've seen on truck are now in process for customers.

We have a backlog of 300 modules for going into 2012 for less than a dozen customers. So the point of it is that this is in our view where the data center is going, is that we have to pay. Whether we're right or wrong, who knows? I just hope someone's right. Like I said, I believe in the world that something's right. So at the end of the day, my sense is that we have to solve these problems, and we've solved them very effectively over time with the means and the information we had at the time that made sense.

We now have an opportunity and the people in this room specifically have an opportunity to define how this is set up so that technology is not limited by the data center and we're not holding back application development, commerce, all of these things that happen higher up in the stack, because we can't deliver data center fast enough or at an economic horizon that makes sense, and more importantly is that second part. Fifteen million, twenty million, fifty, whatever the number is, five million dollars per megawatt isn't going to work. Period.