Building the Green Data Center

By James Turner
June 20, 2008 | Comments: 2

You may also download this file. Running time: 00:21:53

James Turner: This is James Turner for O'Reilly Media. I'm talking today with Bill Coleman, founder, Chairman and CEO of Cassatt Corporation. Before starting Cassatt, Bill co-founded web application powerhouse, BEA. Bill will be keynoting O'Reilly's Velocity Conference this month discussing green data centers. Good day, Bill.

Bill Coleman: How are you?

JT: Doing well. So why don't you start by talking a little bit about how you came from BEA, which is best known again as a web applications company to a company like Cassatt, which is very much data center oriented?

head_coleman.jpgBC: Well in some ways both Cassatt and BEA are related in the emergence of what I'm calling the cloud, the internet. I founded BEA on the assumption that Sun's motto--if the network is going to be the computer then they need an operating system so applications can run on the network. And I founded Cassatt on the opposite, just the inverse of that--that if the networks can be the computer, the network needed an operation system to make the cloud run independent of the applications in the web.

JT: One of the things that has been promoted for Cassatt's, I guess its Collage is the product I've seen mentioned is not only management of the data center, but more efficient use of the data center. Can you talk a little bit about how that works?

BC: Yeah; our focus is really on data center efficiency across the board and we focus right now on three different aspects. Actually the easiest is energy efficiency. You have to do the least about applications and then one step up is application availability and then finally, it's resource repurposing. So it's the focus on making data centers efficient actually also makes them more effective and both cost-effectively and in being available.

JT: What is un-green about a modern data center?

BC: Well data centers were never developed or did not evolve with any thought to power efficiency. You know power wasn't really a consideration as we started in the '60s, '70s, '80s, and '90s and then as the last 15 years when we put networks into the data centers and lots of Unix and Microsoft servers that scaled out like crazy power just went up dramatically, and Moore's Law just put that on steroids by increasing the speed, which increased the heat. And then when we started putting multiple processors and putting them on blades, we just sent this through the roof. I mean the power density; the amount of power per square foot needed to run the data center has gone up about six to one in the last 15 years. In the meantime, we're generating all that heat; we've got to dissipate it, so we run air-conditioning. It turns out the servers only actually use about a third of the power in a data center. Most of it is used to cool it, distribute it, manage the networks, etcetera; so we've just gone through the roof. We had no consideration; it's only been the last five years when two things happened. One is the realization that we can't continue to use unlimited power on this planet and hope to not have global warming change our whole ability to have a good quality of life. And the second is the astronomical cost of power. But today a data center, traditional data center, is probably using eight to nine times the power than it actually could at its optimum most efficient point.

JT: You've broken it down a little bit as far as where the power is going and that that's one of the major impacts. But are there any other environmental impacts of data centers beyond just the power usage?

BC: Well I mean the power usage is the first environmental impact obviously because that power is being generated and unless you happen to be lucky enough to be able to put it next to the Columbia River or such a source it's probably being done by coal-burning plants at least in this country and in China and in these biggest areas, which themselves put out, a one giga-watt power plant puts out 10,000 metric tons a day of CO2. And that is by far the biggest source of environmental impact.

JT: All right; so you mentioned, I think you said it was that a modern data center is using, typically using eight times the power it needs to. Where is the waste coming from?

BC: Well it's coming from--the waste is coming from multiple dimensions. The first is the equipment is nowhere near as efficient as it could be, and thanks to the work in the last five years or so by the, you know the semi-conductor companies, the micro-processor companies, the server manufacturers, the storage manufacturers, there's a lot of work that's been going into that area that; so that's number one.

Number two is the way we cool data centers. Traditionally we didn't think to turn the cooling down when you didn't need it, so once you turned it on you turned on the chiller and it ran forever, and so these things can be a lot cooler than they needed to be. So we've gotten--we're starting to get a lot smarter in how those things are staged and how data centers are laid out.

The third area where really most of the efficiency is lost is that we are way over built in the data center, both because of reliability where you end up with redundant servers for high availability, redundant servers for your tests, redundant servers once again for your disaster recovery; and finally, when we provision those we provision them for what we believe are going to be the peak of all time. So an average data center that is not consolidated or virtualized is running at about an average of five-percent efficiency meaning that 95-percent of the electricity that's being used is being used just to keep idle servers running and to cool--and to keep the cooling going. That's where all that inefficiency is coming from on a traditional data center.

JT: Obviously, one solution that's being suggested to deal with the loading issues is virtualization. Can you address a little bit how virtualization factors in?

BC: Well virtualization is a very good technology. It is part of the data center consolidation. So if you could consolidate your data centers to make them more efficient, to only use the number of servers you need to, and then further you could take applications that currently run on separate computers and put them on one, so if you have 10 computers that are only running five-percent why don't you take two computers and run two of them at 25-percent and you even have some backup capability? So that allows you to raise that utilization level, so that is a very good start, a very good start.

With virtualization today we can easily raise utilization in a high-end data center from five-percent to oh the low double digits; 20-percent is the best number I've heard for a large data center, but the limitations here of course are unless you're willing to mix and match your tests and your batch, jobs like we do on mainframes on the one hand, and knowing that lots of applications don't run on less than a computer; they have to run on many computers--scale out applications, clustered databases, large ERP programs; the virtualization only goes so far.

JT: Where do you see technologies like force.com, S3, Google's new application offering, fitting in? It strikes me that the more you pile together, the more efficiency you can get, because the loads tend to even out the bigger it is. Do you think that that's a direction that improves the situation?

BC: Oh I think it improves it dramatically. So I think this goes in stages; the first stage is what we're doing now in data centers, which is data center consolidation of virtualization. And that takes you to a good level of utilization but you're still going to use three to four times more power on average than you actually need. That's what we call passive power savings. The next stage is to start doing things actively; in other words, being able to know how to move applications around and link them and turn off the servers that aren't necessary dynamically. Then we can really get consolidation way up.

Now if you look at what the Amazon and Google, etcetera, are doing, what they're doing is they're already doing that. They're already sharing servers for lots of loads, so they can keep their utilization up pretty high. I call that Cloud 1.0. It's a very good way of having much higher efficiency applications; however it has a relatively limited application in that it requires your software to be developed on their proprietary systems with their proprietary software. It will only run on those and it can't support today's sophisticated multi-tier applications, which is the basis of most major data centers. So for things that face the web and are one or small tier applications that you don't mind developing from scratch, this is what I call Cloud 1.0. I think we're going to go through a couple of generations until we can turn it into a general purpose utility or what I call Cloud 3.0, which is what--the market we're targeted at.

JT: I've seen online that you--there's a few myths that you like to debunk, so why don't we go through a couple of them? Does data center really need to be 68 or 72-degrees? Does it really hurt to turn computers on and off?

BC: Well I think the myth about turning computers on and off I like to joke to say you know that goes all the way back to when they used to have tubes in them. You know we were really afraid to turn them off. But joking aside, it got really complicated in the early '90s when we started having multiple tier applications and then scaling those applications out on multiple servers on each tier, because now you have to actually know how applications interact with each other to be able to turn them off. But once you know that there's no reason not to turn servers on and off. [Laughs] I mean you turn your PC on and off all the time and the server is no more complicated than a PC; it's actually less because they're--most of them don't have a disk or a display associated with them.

But, so my myth on that is you know if you want to save energy, your mom taught you to turn off the light when you leave the room; well turn off the server when you're not using it. [Laughs]


Velocity is the new O'Reilly conference for people building at Internet scale, happening on June 23-24, 2008 at the San Francisco Airport Marriott in Burlingame, California. Register today!



JT: So there's--it's not like a car where the worst five minutes is the first five minutes when you turn it over?

BC: Well the answer is no; it's not like a car in that regard because they're--we're not talking about a mechanical device. But in the--as in all electronics, you know heat cycles are what ultimately will cause wear. And, you know admittedly, when you get to a few hundred thousand of these heat cycles, you probably will cause wear, but you probably won't be using the server in that--by the time it's gone through that many cycles. [Laughs]

JT: Right; you want to replace it anyway.

BC: You're going to be--yeah, Moore's Law will have made it so obsolete you'll be putting it in your own personal computer museum.

JT: Do you happen to know is 72 or 68 or whatever, sweater requiring temperatures that they run data centers in really an optimal temperature?

BC: I think we tend to run our data centers a lot cooler than they need to be, one because we just don't turn them off, turn off the cooling, and two, because these--a lot of this--these standards go back to when there were a lot of mechanical and things inside the data center and--. But we do run them a little--I think on average most data centers are running too cool; I think the primary reason is we just don't really actually regulate the temperature and we just let it keep getting lower even though at low utilization times servers do use less power. They'll use as low as 30-percent of their full run time power; it's just that's the heat generated from the processor, etcetera.

JT: Another factor might be that in order to get a server rack to be 90-degrees you might have to keep your outside air to be 65-degrees. What are the factors involved in just efficiently using a data center in terms of air flow?

BC: The good news here is in the area of passive power management data consolidation, data center design, more efficient equipment, virtualization; we've come a long way and we have some very good practices in companies like HP and IBM to do that. And what they've learned is how to layout a data center and how to design the air flows to insure that you don't have to keep the ambient temperature at 68 to insure that the server itself is below 80. So part of the whole data center consolidation move isn't just to consolidate into fewer servers and more efficient servers, but also a more efficient layout of the data center, more efficient design of the air flows, more efficient design of the HVAC systems, and that's come dramatically better--gotten dramatically better in the last five or six years. Companies like Rackable have sort of demonstrated how you can optimize that and everybody else is really working that into their designs.

JT: Obviously being green is a good thing, but for a company looking at this issue, how much of the energy costs factor into the operational costs for a data center these days, the operational costs for a computer centric company?

BC: Well the data center costs for electricity have come up dramatically. They've come up from single digits in the beginning of this decade to about an average of 30-percent so they are now the number two or three expense. They're almost at a tie with the actual cost of the equipment; the only cost that's more is the labor costs in a data center. By the way, those are growing much faster than any other costs. As a matter of fact, the overall utilization according to the EPA of data--of power in this country by data centers has doubled as a percentage of power used in the last five years. So even though power usage is going up across the United States, this has doubled as its percentage and they expect in the next five years it will almost double again, so it has become a big cost issue.

Now the irony is most companies still don't measure it against the IT Department. It's still part of facilities, so the IT folks don't actually have it in their budget and that makes it harder for them to actually solve the problem because if they have to pay money to somebody to solve the problem of lower power it doesn't help their budget.

JT: Right; facilities is very happy but--.

BC: [Laughs] Yes; Facilities is happy but the IT folks don't get any credit for it. It's really interesting; we talked to I'll say a Fortune 20, really, really big company, who is now becoming very-known for being green. And one of my Board members talked to their CIO and asked him what he's doing about it, and he said nothing. And, my Board member asked him but wait a minute. Your company is all about green. And he said, yeah; but I'm not--but I'm measured on keeping these systems going--not on saving money for power. When they start giving me a budget for power I'll work on that.

JT: So are there any lessons out of the data center that would be applicable for an SMB or even for a personal user in terms of ways they can reduce their costs and help the environment?

BC: Well I mean I think the first lesson for an individual user is just to turn it off when you're not using it. I mean it's not going to hurt anything. People--every--most people are migrating to notebooks anyway and you carry them on and turn them on and off; that's actually no big deal. But the lesson for SMB is the market for software as a service is SMB; that's the future of software as a service. We--we're a relatively small company. We have about 100 employees, but we have people in about 14 locations. And we happen to use NetSuite for all of our business software; we use an online travel aid company that relates to that; we use a voice over IP so everybody anywhere is connected and yet we don't have to run any of that software ourselves. That's not only much more efficient from a power point of view; it's certainly much more efficient from our point of view. We don't have to actually spend all that money or use all that power. I think that's the future of small to medium-sized business.

JT: You're going to be speaking, keynoting in fact, at Velocity, O'Reilly's Data Center Conference coming up. What are you going to be talking about?

BC: Well I'm going to talk about the velocity of the data center itself. I'm going to go back; I'm going to start by going back and saying hey, look at how simple a data center was 15 years ago, and look at what we've done to it. The scale and complexity have gone through the roof; they interact with each other to the point you can't do anything. We have so many applications so tied together in these big data centers on servers, we don't even know where they are; we won't turn them off. The power is through the roof. The efficiency is really low. As soon as you run out of power you spend $100 million to buy another one. So what can we do?

And what I'm going to say is we can start right today with energy efficiency. We need to do the passive part, consolidate and do that well; we need to do the active part. But as soon as you do the active part, which is what we are trying to sell, which is the ability to dynamically measure your applications and determine how to scale the power on and off based on the demand and the need, you become much more efficient in your operations. But as soon as you do that, the light-bulb is going to go off, because once you know that you can turn applications on and off and move them around, you can now begin to build into a full cloud yourself. And as soon as you can start building into a full cloud yourself, you can start outsourcing that cloud and letting other people run that capacity and that's when we're going to start getting real efficiencies not just in operation and reliability, but in utilization and power across the whole industry. That's what I'm going to try to say.

JT: All right; we've been speaking with Bill Coleman, founder, Chairman and CEO of Cassatt Corporation. You can hear more from Bill at O'Reilly's Velocity Conference coming up this month. Thank you for talking to us, Bill.

BC: Thank you; it was my pleasure and I'm looking forward to the Conference next Monday.


You might also be interested in:


2 Comments

We also considered server room location in our design. placing it just above the car park level, which has been more energy insensive to heat. We use the waste heat from the data centre to help heat the rest of the floor. Excess heat in summer is then just vented to the car park.

This is an extremely long but very interesting read.

We also considered server room location in our design. placing it just above the car park level, which has been more energy insensive to heat. We use the waste heat from the data centre to help heat the rest of the floor. Excess heat in summer is then just vented to the car park. This is a good idea,I will also go to make one.

I like your book and the article, I have also learned many things from inside. I work in kswchina, a examinations site about education. I am very happy that I can know so many experts in here. My name is stone wang .

Popular Topics

Archives

Or, visit our complete archives.

Recommended for You

Got a Question?