Transcript of Green HPC Episode 2: IT, HPC, and where the twain shall meet
This is the transcript for episode two of the Green HPC podcast series, IT, HPC, and where the twain shall meet. You can find out more about the series at the Green HPC podcast series home page, and you can listen to the audio of this episode, find out more about the speakers, and get access to links and presentations that they’ve suggested at the episode 2 homepage.
[0:00] insideHPC is a proud sponsor of SCO9, the international conference of high-performance computing, networking, storage, and analysis.
[0:08] Is your data center green, or a snarky shade of brown? If you’re burning up the kilowatts, come to Portland, Oregon, this November for SCO9, and learn from the experts. SCO9: Going strong and going green www.sc09.supercomputing.org.
[0:27] As you have seen from Secretary Chu’s website, and from what President Obama has made clear, that reducing our dependence — or as they say, reducing our addiction — on foreign oil, coming up with clean, renewable energy resources and reducing our consumption is one of our prime goals.
[0:53] That was Pete Beckman at the top of the podcast. Pete is from the leadership computing facility at Argon National Laboratory, a group I’ve been following a lot over the past year, as I really started to focus on the growing energy awareness at HPC. I’m John West, from insidehpc.com.
[1:10] Pete was talking about what the big picture of green HPC looks like for him. Argon is part of the Department of Energy, and for them, they have one of those rare ties between the things that we read about in the newspaper every day that the President and his Cabinet are doing, and the things that Pete’s team does in designing, running, and building their supercomputers.
[1:31] This is Episode Two of the HPC Green Podcast Series from insideHPC.com. And today we are going to be talking about the differences between going green in general IT, and going green at HPC.
[1:43] We’ll hear more from Pete Beckman, and then we’ll talk with Ed Turkel and Steve Cumings of HP about their experiences on both sides of the IT business, and learn about how the attitudes and solutions differ in IT and HPC.
[1:57] In our last segment today, we’ll talk with Christian Belady from Microsoft about what the real drivers are for adopting green technologies, and what it’s going to take to see a big shift in that direction.
[2:09] But to get things started, we’re going to talk with Pat Tiernen of the Climate Savers Computing Initiative, an industry group that’s focused on awareness in green IT. I put the question to him that when you dig in to all of the EPA reports and estimates on the fraction of power we use, it’s 1.5% or 2%, and that’s for the whole category of information and communications technology, or ICT.
[2:31] That includes everything from cell phones and network switches, to servers and HPC gear. That’s really not a lot of power in the big scheme of things, so why all of the attention now?
[2:42] Well, as an example, back a year or so, a year and a half ago, when it was first known, let’s say, or made widely known, I should say that the ICT industry was 2% of the emissions problem. That’s on par, if you will, with the airline industry.
[3:05] What happens when it’s 6%, you know? Back then, there was a billion PCs out on the market. There’s a number of research docs out there, and if you just look at the numbers that show that the number of devices will be somewhere around two and a half billion over the next few years.
[3:22] Look at what’s happening with blades. Look at what’s happening with the proliferation of these devices. And you can extend that to things like cell phones. We need to make sure that as these things proliferate, for good reasons and for necessary reasons; they are operating at the highest level possible.
[3:39] So you know that, I think to me, is one of the reasons. One of the other best reasons is, why not? There’s a value of proposition in doing so, it can return value to your shareholders, it can make a contribution to help slow the effects of climate change.
[3:56] So those two things together to me make it compelling. Now if you look out a little bit ahead, and you look at what’s happening from a regulatory perspective, I don’t think there’s any debate anymore about climate change and the source of emissions and whether we need to act or whether we don’t.
[4:15] Look at the legislative dockets around the world, and it’s coming. Let’s take a look at the CRC in the UK, the Carbon Reduction Commitment Program. The top 10,000 businesses in the UK, regardless of whether they’re participating in the European ETS Scheme or not, is going to have to report on this.
[4:33] Regardless of whether you’re directly regulated in the next couple of years from this, it’s probably a good guess that in one form or another, you’ll be impacted, and so I think it’s a great motivator to get engaged. Start the learning; and why wouldn’t you anyway, if it improves your business anyway?
[4:54] So the dramatic growth of ICT at all levels means that the fraction of power we use will continue to grow. And by starting to manage that growth now, we have a chance to get ahead of a much bigger problem, which makes sense I guess, for ICT as a whole, but what about for HPC?
[5:10] We’ve talked about the energy cost issues for individual centers, and those are very relevant to those centers. But as we heard last time from Horst Simon, energy reduction as a cost savings doesn’t need to be connected with helping the environment by reducing carbon footprint. It can be, and at least some of the conversation about green and HPC is around this issue.
[5:31] So I put it to Pete Beckman. If total energy use and ICT is only two percent of the total, and that two percent is really a pretty small number, but the fraction of energy use in ICT that goes to HPC is even smaller. Are we big enough to even matter?
[5:49] That’s a great question. Yes, we do have enough resources that we’re consuming to make this very efficient, and to want to look at reducing the power. Our data center, for example, our supercomputer takes 1.2 megawatts of power.
[6:06] Now in the grand scheme of things, compared to running Google and Yahoo across the planet, well these are, as you point out, relatively small. However when you compare and add up all of the technology and high performance computing across the nation, it’s still pretty substantial.
[6:23] Furthermore, the technology that you see in computing, in high performance computing, in scientific computing, is really looking into the future. What we are able to deploy now are the kinds of things that people will deploy on their desktop years from now.
[6:41] So if we can get it right in terms of energy and power and utilization and reducing our footprint at supercomputer level, we’ll be getting it right all the way down to the desktop.
[6:55] Which all makes sense in theory, but how do actual customers behave? For that, I went to a company that is prominent on both sides of the IT equation, HP. We talked with Ed Turkel, and asked him what attitudes motivate customers that come to them looking to reduce energy consumption. Are they coming in with an environmental mindset, or are they just looking to cut costs?
[7:15] Well, I think it’s different with different customer segments, actually. Because one of the things that we’ve been seeing, and again, when we talk to customers in DOE as an interesting example is, there’s a set of customers where again, when we’re talking about a request for a new proposal for a new HPC cluster or something like that, the HPC folks are looking at total cost of ownership.
[7:40] So they are including power, calculations, and so on and so forth, in their total costing. In the case that I talked about before, that’s the kind of discussion where we have, factoring in the power footprint, and so on.
[7:56] There’s also a class of customers where the infrastructure, the data center, and power supply and things like that is simply managed by different folks. That, by the way, tends to be not uncommon in some of the government agencies.
[8:10] So in those cases, you’ll get the folks who are actually deploying the clusters who are not per se looking at the cost of power, but they are being mandated to think green. And so it can be both ways. You know again, where a simple policy of trying to be green is the motivating factor in other cases. Again, it’s just the simple cost of power that’s factoring into a total cost of ownership.
[8:42] So when you talk with customers, do you have a sense that it’s the same division on the Enterprise and HPC side, or does one of them care more about energy issues than the other?
[8:52] Yeah, I’m not sure I could quote a clear percentage there, but I would say in my experience on the enterprise side is that they tend to be more focused on total cost of ownership, and actually to be responsible for the power cost as well as the cost of the solution itself.
[9:11] But again, there are cases where that’s simply not true. Again, depending on how they’re organized, and who manages facilities versus who manages the data center, and things like that, can impact all of that.
[9:25] We’ve done some research in terms of how customers make buying decisions you’d expect, but also, specifically, how they make buying decisions in terms of energy also.
[9:34] That’s Steve Cumings, also with HP.
[9:37] So the answer a year ago, I’ll just put it that way, is that power and energy consumption was becoming a purchase decision. So in other words, it’s typically not the primary person’s decision, but it’s in the top three, so it’s one of the things that you have to meet for their purchases.
[9:55] And you know, we were seeing that percentage sort of reach at least halfway in the kind of 50% range for an Enterprise customer. And that’s not saying that they don’t care about the environment, it’s just that these guys also have to translate that into money. And as Ed said, it’s a TCO calculation.
[10:14] And my guess is, with the economy the way it is, it very much increases the same way, so we are seeing increased focus and interest and inclusion in purchasing decisions, in terms of energy. I would say on the Scale-Out on the HPC side, it’s probably higher than that, simply because these guys are running such large IT footprints. So they care that much more about how many watts each node is burning.
[10:37] Do you see different kinds of solutions applied to Enterprise computing as opposed to more traditional HPC kinds of deployments?
[10:44] So, not a hard and fast line, but I would say that you do see a difference in behavior, because the Enterprise customer is very interested in energy, but they’re also interested in all of the same features that they’re used to.
[10:54] So they’re looking for more of a general purpose type environment, where they have the management tools that they’re used to, and all of the redundancy that they’re used to, and things like that. So they’re interested in, "I want you to take all of the energy possible out of the servers that I’m used to buying, because I like what I’m buying, but I need it to be more energy efficient."
[11:13] Scale-Out in HPC type customer will make specific choices and can, because they have a very focused environment. And they’ll actually go beyond that and say, "No, I want you to deliver a purpose-built server. I’m willing to avoid these kind of components, because I don’t need them. And with that in mind, I expect you to be hyper-efficient."
[11:32] So it’s a question, I think, of focus. The Scale-Out in HPC guys are deciding, "I need an IT footprint that does this, and I don’t want to spend power or money on anything else." Versus the enterprise customer is looking at a data center that they’re used to, and the type of IT that they’re used to buying, and want you to optimize their energy footprint for that type of environment.
[11:56] We should also recognize that HPC isn’t only Extreme Scale-Out, that there’s a range of requirements at HPC. So there’s a piece of HPC that looks a lot like enterprise. So they have some of the same mission-critical kind of requirements, and the same expectation of having a similar level of management that they’ve come to know and love.
[12:18] And by the way, those are the folks who are likely to be continuing to use blades for HPC, because that environment provides very much that kind of rich management capability, and so on. And you know, we start getting to the other side of HPC that’s a little bit more Extreme Scale-Out if you will, and so on, that the newer capabilities are going to be valued.
[12:44] One of the things I see is, I read about these technologies, and report on these changes in the industry around energy awareness, is that adoption seems to be pretty variable. In general, I have a sense that the enterprise side is ahead of where we are at HPC.
[12:59] So I got to wondering why that is, and what it will take to accelerate the pace of adoption at HPC, especially given the non-environmental reasons that we have at large scale, to start reducing energy use.
[13:10] To me, it’s a differentiator.
[13:12] Christian Belady, the principal power and cooling architect with Microsoft’s Global Foundation services infrastructure team.
[13:19] And as long as you’re operating in a business where there’s competition, that’s where adoption will happen first. Within HPC, it’s not driven by competition; it may be slow to adopt. But ultimately,I think when people start looking at, "How do I get the most computation for the lowest possible cost?" And if it’s done in some competitive environment, the person who did the more computation for the lower cost will win.
[13:44] Yeah, so from our perspective, it really comes down to the top-line budgets that we get from our funding organizations being essentially fixed, and if we want to continue to grow computation at the rate that we’ve grown it in the past, then we are going to have to do that by managing the infrastructure costs.
[13:59] Some people can now buy more machines than they can afford to power. Or even more common that they have, you know, copper and transformers and UPSs to run. And in a federal facility like mine, you can go to the Water and Energy Resources Act for a meter of last year and the year before, and I actually have a line item, a literal act of Congress to give me the authority to spend money I already had, so it wasn’t an add. I just needed the authority to spend money I already had to build out a new footprint, and the power and the chillers, and all of that kind of stuff. I think that’s going to be a more powerful motivation for us.
[14:38] Exactly, and was it Edison that said, "Innovation only occurs with constraints?"
[14:44] And competition drives constraints, because you have to be more profitable. If the budgets get smaller, and you have to have bigger solutions, you have to innovate, right? And I think that’s the game here, and so HPC I guess will have the same opportunities.
[15:02] That’s it for this episode of the Green HPC Podcast Series. You can find out more about the topics and people in this episode by going to insidehpc.com and clicking on the link for the Green HPC Podcast Series.
[15:15] On the next episode, we’ll talk about the specific benefits to your HPC center of adopting a more energy-aware posture. We’ll talk with some of the leading vendors in this area, including IBM, Invidia, and Cray, and we’ll also hear from the team building one of the petaflop scale computing systems coming online soon, about why they had no choice but to think green even before they started pouring the foundation for their new data center.
[15:39] Until then, I’m John West. For all of us here at insidehpc.com, thanks for listening.
[15:46] Hi, I’m John Kirkley, Science and Technology writer for the HPC community. If you need white papers, case studies, or other marketing collateral, I’m your man. And don’t forget podcasts. Check out my website at www.kirkleycommunications.com.