June 2011 Webinar Playback:
Key Data Center Metrics and How they are Applied
In today’s dynamic business environment the need for more capacity in the data center continues to increase at a rapid pace. While the cost of energy is also on the rise, business leaders and corporations are feeling the economic pressure when addressing these needs in the data center. These leaders, along with global public policy makers also recognize energy consumption as a strategic and critical issue.
Focused on energy efficiency, many data center managers and operators have been discussing and implementing energy efficiency metrics. Yet these metrics are not always applied clearly and consistently across the industry.
Our panel of experts; Mark Monroe, Executive Director of The Green Grid, Scot Heath, CTO at 42U and John Pflueger, Principal Environmental Strategist at Dell, will discuss key energy efficiency metrics, their practical applications in the data center and shed light on why you should care about these metrics. Learn how you can use these metrics for planning, setting objectives and reducing energy consumption in your data center.
This Webinar covers the following data center metrics toipics:
- Overview of Data Center Energy Efficiency Metrics
- Practical application of Data Center Metrics
- Who uses these Metrics?
- Worldwide Trends and the Future
- Q & A
Read the Full Transcription
Rebecca Franco: Ladies and gentlemen, thanks for standing by and welcome to today’s session in this 2011 42U Web seminar series.
My name is Rebecca Franco and I’ll be your moderator for today’s webinar, Key Data Center Metrics, Practical Applications, and the Way Forward.
During this presentation, all participants will be in a listen-only mode. However, we encourage your questions or comments at any time through the chat feature located at the lower left of your screen. These questions will be address as time allows.
This webinar is being recorded today, June 28, 2011. A replay of this webinar recording will be available on our Web site 42U.com approximately 48 hours after our presentation.
Our panelists for today’s presentation include, The Green Grids’ Executive Director, Mark Monroe, 42U’s Chief Technology Officer, Scot Heath, and Dell’s Principal Environmental Strategist and Board Member of The Green Grid, John Pflueger.
At this time, I’d like to turn the presentation over to Mark.
Mark Monroe: Thanks very much. Well, thank you everyone for joining in. We wanted to talk today about metrics that some at The Green Grid has published and the importance of metrics in general. And I’ll kick off the discussion with starting out on some of that and talk about the lifecycle of a metric, and then we’ll get into some of the metrics that The Green Grid has proposed using the power metrics and efficiency metrics.
So first, let’s kind of discuss the need for metrics overall before we get into the slides, and it kind of doesn’t matter, the phrase has been attributed to different people, whether it’s Lord Kelvin or Peter Drucker or Dr. Deming or – you know, efficiency experts everywhere or measurement experts everywhere say frequently, “You can’t manage what you can’t measure.”
And the importance of metrics is that it gives us a way to quantify the intuitive senses that we have of work being done or energy being used, and you can see how important that is in some of the progress that’s happened. If you don’t have metrics that are used to drive behavior, things will happen at random.
In 2005, back – if we step back to 2005 there was really no measure of data center efficiency. The field was wide open because most of the people didn’t even think about data center efficiency. This was before the economic downturn, before $100 a barrel oil, and the green business wave was just beginning to start.
What most people focused on was the fairly recently announced at that point Uptime tier classifications that talked about reliability of data centers, and so this was a kind of a metric, a short hand way of describing the reliability of the design of the data center, and that metric was one of the ones that people focused on. And you may be familiar with these – the Uptime tier classifications, in terms of Tier 1, 2, 3, 4 reliability.
It wasn’t that people had not built data centers before the classification existed, but it was that we didn’t have the short hand wave describing them and measuring whether they felt into that classification or not.
Around 2006, a couple of technicians from Hewlett Packard, Christian Belady and Chris Malone, developed the – a measure of efficiency of the data center. This was now getting into the point where the green business wave was starting to crest and folks started to wonder, “Geez, what is the – where does the energy in a data center go? And we’ve kind of got the reliability thing down, but we need some kind of a measure of how much energy we’re using in the data center and where it goes.”
The first definitions of the metric came out early and The Green Grid latched onto this as Christian Belady and Chris Malone’s companies were both members of The Green Grid, and started to publish this metric. Without the metric, things were all over the place.
An example that I can give you is that Lawrence Berkeley, National Labs, in 2006, shortly after the power usage effectiveness metric had been proposed by Belady and Malone, did a study and came out with an average in a small number of data centers, but they said the average measure of that PUE was 2.2.
A recent study by the Uptime Institute, that PUE has now been measured for four of five years and brought along in its maturity by The Green Grid, said that a survey by the Uptime Institute of 500 large companies said that they’re average PUE was down to 1.8.
Again, the importance of that metric, that 33% drop from a random survey done by Lawrence Berkeley in 2006 to a 2011 survey done in – by Uptime, it could be worth somewhere between $300,000 to $500,000 per year per megawatt of capacity in a data center’s operating budget. So, you can start to see where the importance of metric gets changed, not only from an efficiency standpoint and a green business standpoint, but from an economic standpoint, which is where the rubber hits the road I most metric definitions as well.
Without the metrics, we really don’t know where we are. There’s another survey done by Green Grid member of Digital Realty Trust that show as recently as February of 2011, 20% – nearly 20% of the executives in large companies, and these are defined by digital as companies with over $1 billion in revenue or over 5000 employees, don’t know what their power usage effectiveness is, and without knowing, they have no way of managing it. So, they’re probably on the high end and wasting money and not understanding that they can be saving money and running their business more efficiently through the use of the metric.
There’s – and power usage effectiveness is just one aspect of the sustainability metrics that we in The Green Grid think are important. That obviously deals with energy. The Green Grid has also announced some metrics that deal with carbon and with water, and in the future you may see things about electronic waste or materials or, you know, all kinds of aspects of sustainability that haven’t existed in a standardized way before.
So as we think about those metrics, now might be a good time, let’s go ahead and go to the next slide, please, about the lifecycle of a metric.
Metrics, as I said, the PUE metric, let’s take and follow it through this kind of a chart, if they go – if they have a lifecycle where things are conceived and in the beginning of the inception process, talked about early on, communicated to a broader audience where people start to think about the metric and how they might use it.
And they make some initial trial measurements and, you know, we end up with things like PUEs of less than 1.0, which means that your data center is actually producing energy rather than consuming it, and I sure want to meet the people that are having perpetual energy data centers out there. But, through that process of communication back and forth between the developers of the metric and the users of the metric, you refine that definition.
And what – and The Green Grid’s standardization or The Green Grid’s PUE model has gone through this for some years now and is starting to become quite refined in its definition. The acceptance of this in a wide range comes in over time as well, and again the – I point to the Digital Realty Trust survey from February that says that 80% of people do know what their PUE is and are working towards improving it.
And then finally, the step where everyone begins to do things in the same way, you know, we might call standardization and this might be where you took the definition – the widely accepted definition of a metric, how it’s measured, how it’s reported, how the data is collected, and turn it into an actual international standard.
And there are lots of standards bodies out there available, the ANC Organization in the United States, ISO and IEC in the international communities, where they would now take a very rigorous approach to defining and how the metric was measured and reported and what kind of things it could be used for and carry it through.
And so, PUE again is a good example. I would say the PUE today is in the acceptance phase of this diagram where it has been defined and communicated and refined over time. The global harmonization efforts that have happened over the course of PUE’s lifecycle, and I think John will talk about those more, have led us to a highly reliable reporting of power usage effectiveness, and very useful reporting of that as well.
And we see other metrics that The Green Grid has proposed, just in the inception phase. So, for example, carbon usage effectiveness was proposed in December of 2010, and water usage effectiveness was announced as a data center metric – a proposed data center metric back in the March timeframe of 2011.
There have been little – a little bit of take up on that. I’ve been to some conferences where people have talked about those metrics and are starting to use them and starting to report them, so we’re just barely moving in to the communication phase on those, and I’d expect that they’d have a similar lifecycle like PUE.
So, to talk about PUE and some of the other metrics or the other sustainability metrics, I’m going to turn it over to John, who I think is going to go to the next slide.
John Pflueger: Thank you very much, Mark. So, Mark mentioned a few of these things a moment ago. When we were planning out the webinar and trying to figure out, you know, what we felt it was important to communicate to the audience, one of the things that came out with – that, you know, we’re interested in promoting PUE, and one of the ways in which we can do that is to provide some – you know, provide some background on how well developed it actually is. And I think it segue’s very well from the model that Mark presented just a moment ago, in terms of the lifecycle of a metric.
As Mark had mentioned, the PUE predates the public announcement of The Green Grid. The industry had been really looking for something to help it understand the efficiency with which power was being distributed to the IT equipment, and there were a number of various proposals for doing that.
And really, you know, even before the public announcement of The Green Grid, the founding members had gotten together and said, “You know what, we think we can really get behind the PUE thing. It feels simple enough; it feels directionally correct; we believe that it’s going to drive the right behavior.”
And so, when we came out of the gate with The Green Grid in February 2007, really the first White Papers we published talked about PUE and that Green Grid belief that PUE and DCIE, which is just another version of PUE, were really, you know, what we wanted to get started with, and so we started there.
And then, we had to tackle the problem of figuring out, you know, “How do we actually collect the data? How do we take the numbers? How do we put the metric into practice?” And a number of organizations did so and as we went through those learning cycles and we started understanding more and more about what we were going – what guidelines we were going to need to lay down, so that the broader industry would really be able to make use of the metric.
And about, I’d say, about a year after the initial White Papers came out we were able to really put out some fairly detailed information on how to – you know, how to take the data, how to calculate PUE, and even how to calculate different levels of PUE based on how good your own internal data collection opportunities were.
Some folks are highly instrumented, some folks not so much, and we needed a way to really differentiate reports coming from facilities that were highly instrumented with reports where folks really wanted to collect a metric, but had to do it with sneakers and notebooks.
Over time, with more and more experience with PUE, we were able to provide additional detail, really refine a nomenclature for PUE so that we could understand, you know, for any given report how the information was calculated, what it meant, and really providing guidelines and ground rules for the industry to follow as it used this metric.
As more and more people began to adopt PUE, as more and more data centers started using it as a tool to manage their physical infrastructure, we started seeing a lot of interest globally and a lot of interest among a lot of different organizations in really making sure that not only was this a Green Grid thing, but this was something that all the major interested stakeholders in the industry could get behind.
And so, we started with a number of other key entities, a global harmonization project, which included folks such as 7×24 Exchange, the Uptime Institute, certainly The Green Grid, ASHRAE, the Department of Energy’s Save Energy Now program, the folks at the EPA and Energy Star, the U.S. – I think the U.S. Green Building Council was involved, Silicon Valley Leadership Group, and those were just the U.S. interests.
The global interests were the Green IT Promotional Council in Japan, European Commission Joint Research Center for the Data Center Code of Conduct. And really, the group worked together on a difficult task. It’s a simple metric, but it is challenging to get everybody to achieve consensus, and as a result of that work we made a couple of changes.
We came back out with some new guidance and we have – we now have something that worldwide is accepted and recognized across the globe and across the industry. And this helps the industry immensely because as a given data center or a given organization really wants to consider adoption of this metric, they want to know as well there’s some weight behind it, there are folks who are watching over it and making sure that it’s as strong and robust a metric as possible.
Since the global harmonization work and I believe that parties continue to talk because there’s always more work to do, as Mark had mentioned we now have some offshoots of PUE. The Green Grid found itself in the last half of last year really being drawn into sustainability issues that are adjacent to energy efficiency, other resources that are either generated by or consumed by the data center; notably, carbon emissions and water.
And it was fairly easy for The Green Grid to come out with some offshoots from PUE that helped data centers begin the process of understanding their greater impact, in terms of resource – you know, resource consumption and resource usage and possible resource wastage.
So I think with that, and kind of after talking about the timeline a little bit, I’m going to pass you over to Scot and Scot’s going to provide a little bit more detail on PUE and partial PUE and how you might apply it within your facilities.
Scot Heath: Thank you, John. So as John mentioned, you know, many, many organizations, most of which we have listed here, are U.S. organizations, in fact they all are, have worked diligently to standardize the way that we apply PUE. And in fact, there’s a paper published very recently, May of this year, which is Version 2 of the PUE guidelines for measuring and reporting.
And this particular guideline, you know, is a simplification, in my opinion, of the last guideline. The reporting structure is a little bit less complicated, but it really lays down the law, if you will, about you know how accurate or how precise we want to be in gathering information to make it useful. And the answer is you don’t have to be all that accurate and all that precise to have useful information. What you do have to do is be consistent about the way you gather it, and be consistent about the way you report it.
Now, we don’t, you know, necessarily encourage, in fact we discourage, comparing PUE between organizations. So, you heard John and Mark both talk about averages, you know, where you’ve seen the average PUE reported go from 2.2 down to 1.8. I in fact, during my time at HP, we had a group that did this and our average about two years ago was exactly 2.0, so it fits very nicely on that timeline of people improving PUE over time.
And that’s kind of a bulk measure and everyone on this call probably understands that, you know, the absolute function of your data center, the structure of the data center, playing in with those tiers that Mark mentioned, vastly affect your ability to have a PUE that is low, and we don’t want that to be a deterrent from making improvements on your PUE. So, the bottom line of all that rambling is that, you know, we highly encourage you to measure this and then make progress against your measure.
Now the measurement part, as John mentioned, it’s a very simple metric, right? The – so the metric is the total energy of the data center divided by the amount that goes to the IT equipment. And it seems that, you know, intuitively that would be something that would be very easy to go into any one-line diagram and maybe you’ve got gas and maybe you’ve got some other energy source, but you can sum all those things up.
Well, I’ll tell you, I sat in a working session with some of the folks on the committee that were actually developing the measurement guidelines a couple years ago down in St. Louis, and I bet we spent, you know, 30 minutes just talking about how we define the data center. Not necessarily the big data centers, which are you know standalone buildings and clearly all the support in that building goes towards making the data center work, but (mix used) facilities.
You know, 75% of the IT equipment lives in data centers that are approximately 10,000 square feet or less, and there are a lot of those out there. And, you know, many of those have got some technology that presents some of the biggest challenges to have good PUE numbers. So, those are the ones that we really want to encourage, you know, instrumenting and looking at these metrics.
So calculating PUE, you know, here are some basic guidelines. IT energy consumption should be a minimum measure at the UPS, but you know the fact of the matter is if you can’t measure it at the UPS, we still encourage you to go, you know, estimate what that is.
And in fact, the new guidelines have four different levels, if you will, of PUE precision and the first one is the demand-based measurement. That is, you know, I’m just going to go look at watts and I may read that off the UPS. I may time meter. I may get that from a variety of methods, but the kind of common theme amongst all these different levels of precision is I want to do it for an extended period of time.
You know, just a few weeks ago, The Green Grid presented a study by some of its members who go together and actually did an efficiency upgrade in an active data center with a goal of improving – they let PUE improvement be their guiding factor. And I remember hearing this at the conference in February that, you know, it’s a very difficult thing to do on a short-term basis.
And the reason for that is, you know, daily activity, daily changes in the environments outside caused significant swings in instantaneous PUE. And so, you know, taking PUE measurements over a short period of time is basically not in the definition of that Version 2 guideline.
So, we can look at demand meters. A better measure is to look at usage. And so, going to Level 1, Level 1 through Level 3 are actually based on usage. What’s the total energy consumed, and not the rate at which I’m consuming it? And really the difference between Levels 1, 2, and 3 are how fine a line do we cut getting to the IT equipment?
Level 1 is measuring the output of the – or measuring the IT power at the output of the UPS. Typically, a very easy thing to do, most – at least building level UPS is – (floor level) UPS’s have got, you know, built in metering, you can walk right up to the panel and read that off. You know, as John noted, sneakernet, you walk around with a notebook and write that down. If you have some way to integrate that over time and actually get energy use on a fine enough grain basis, then you can turn that from a Level 0 into a Level 1.
Level 2 goes one step farther down the chain and looks at, you know, IT power being the output of the Power Distribution Unit. And the reason we want to do this, of course, is those pieces of equipment as we get closer and closer to the IT equipment are not doing useful work. They are expending energy. You know, anything that’s locked in there gets turned into heat without getting any compute cycles out of it. So, we want to get closer and closer to the piece of IT equipment to increase the precision of the measurement that we make.
And finally to get to Level 3, we’d actually like to measure the power at the input to the IT equipment itself, and so – and there’s a move within The Green Grid. I was Co-Chair of a committee for a while defining the different ways that equipment measure its own internal parameter, all kinds of parameters, and report that in a standardized fashion all in an attempt to make, you know, the application of their various metrics as easy as possible.
And you’ll hear later on that that’s, you know, a theme that we hear loud and clear from members of The Green Grid, but above all we would like to make these metrics quick to apply, you know, without having excess overhead. We have enough to do as it is without gathering information here.
So I think with that, I will turn this back over to John. We’ve talked so far about all kinds of energy efficient, sustainability metrics. As of late, you know, there’s been a move to institute some productivity metrics, not necessarily power, but you know how are doing useful work here? And I’m sure John will have a lot to say about that.
John Pflueger: Thank you, Scot. So, PUE is a very, very good metric. It’s good at what it does. It helps provide guidance for power distribution architectures, and cooling architectures. It doesn’t, and it was really never meant to, provide any guidance for IT equipment.
And if we really want to make continual progress and even more progress in making data centers more energy efficient, we’ve got to have metrics that help us understand and optimize our IT architectures.
So along those lines, for a couple of years now, The Green Grid has been involved in investigating something called Productivity Metrics. Metrics that would look at the data center and try to estimate how much energy is it taking to – in the data center to generate a certain amount of work?
Now I have to mention right off the top, this is a work in progress. This is ongoing. We’ve come up with a number of very, very interesting pieces of work out of it. We don’t have any end conclusions at the moment. But like I said, this is a work in progress.
One of our first minor epiphanies in this is the reason why this has been so hard to do over the past few years is that productivity means different things to the different organizations. I was fortunate enough a few weeks back to be invited onto a panel at the Uptime Institute’s Symposium in Santa Clara and sharing the stage with folks from SAP and Yahoo and Akamai.
And Akamai, (Nicole) from Akamai had come out and said, “Hey, this is what we’re using to measure productivity. We measure watts per unit of bandwidth.” And this made a lot of sense to me because Akamai’s data centers are really factory-like. The data center is something where it’s generating something that Akamai is directly selling.
And companies like Akamai and eBay and Google and Yahoo and Microsoft, where the data center is a factory, have to have internal custom productivity metrics to help them guide their business. It’s part of their way of understanding their cost of product.
And we’re starting to see examples of these show up in the media. I mentioned Akamai. We’ve heard now some public reports about Verizon doing something very similar and AT&T doing something very similar, and I really expect to see more and more of these come out over time as companies start to mature in their measurement of these figures.
Now, that’s great for the factory-like data center, but this only a small subset of all the data centers that are out there. And really, data centers where you have a large number of different applications over the entire server population would find it hard to leverage that approach, because only a few of the servers in the data center are generating Web searches or generating one type of data, and part of the data center is doing calculations and part of the data center is running transactions.
So for this other large class of data centers, we’ve been looking at trying to find proxies. Things that we can measure within the data center that maybe they’re not direct measurements of productivity, but they’re indicators; things that we can tap into or measure that are directionally correct. Not necessarily 100% accurate, but something that we’re – where we can get the information cheaply, quickly, and easily.
And in our own investigations, as we’ve spoken with focus groups about these metrics, the number one thing that they said is, “Look, we need something that’s easy to use. If it’s directionally correct we can sacrifice accuracy as long as it’s something that we can implement.”
So as I mentioned The Green Grid has been working on this for a couple of years now. We’ve got a number of White Papers on the topic. From the very first White Paper when we started investigating a metric called DCeP to a second one where we kind of laid out the proxy program, to a third one where we’ve kind of narrowed down and understood more about the proxies.
And we’re currently working collaboratively with Pacific Northwest National Labs to run an experiment and write up the results where on one set of servers we can run a sample workload and really collect and calculate these different proxies, and look at them relative to each other to see how well they work at establishing or estimating the amount of work being produced by these servers, in terms of, you know, computation or transactions or the real workload happening in the processor.
So, this work is continuing. My recommendation to the industry is really to track what we’re doing, get involved where you can, be aware of what’s going on. And as we further this work start looking within your own organization at what you might be able to measure, and how you might be able to integrate it in to your own management tools and techniques in your organization. We want something that will do for IT equipment what PUE has done for power distribution and cooling architectures.
So, I think the folks at 42U have also arranged one more topic for something that The Green Grid is also very proud of, and in this case it’s not a quantitative metric, but really a large, broad-scaled view of the entire data center.
In February at – actually in March at The Green Grid’s Technical Forum in Santa Clara, we really came out with Version 1 of the Data Center Maturity model. This was a piece of work that really was driven by Harkirat Singh, out of Thomson Reuters in Europe, who had a vision for a way that we could maybe qualitatively assess where we are today against where we might be five years out from now. And help organizations plot a path that will lead them to a more efficient, more productive data center in the future, and not just from a power and cooling perspective, but across the entire data center, whether it’s power or cooling or servers or storage or networking.
We decided to latch upon a model based on a number of levels where, you know, instead of a number like 36.1473859, you know, you’re either Level 0, Level 1, Level 2, something where you could come in and fairly easily assess where you are.
I think I’d like you guys to go forward to the next slide and I want to preface it by saying we know this is an eye chart, so when we put this slide in the presentation it was not with the intent that you guys go read all of this size 1 or size 0 print that’s in here. But, we wanted to give you a feel for the depth of content that’s in the Maturity Model. At the very top you see a number of green arrows leading to the right from light green to dark green, and these are the different levels; Level 0 all the way to Level 5.
We reserved Level 0 for a state where you’re really not doing anything or the barest minimum. Level 1 would be a level where we’d expect most organizations or most facilities to be today. And we set Level 2 for best practice today. So, the intent was that the folks who were absolutely doing the best in a particular area today would not be able to do any better than level 2.
And then, we proceeded to sit back for each one of the areas in the model, you know, there are eight different sections in the Maturity Model, and look at what do we think best practice looks like five years from now, and that’s Level 5. And not only that, but you know we don’t expect folks to be able to hit best practice five years from now, and it may not even be appropriate for an organization to want to be at Level 5. They may say, “You know what, it’s good enough for us if we’re Level 3 or Level 4.”
So, we identified a couple of intermediate steps in between Level 2 and Level 5 where organizations can target, and we went through a fairly exhaustive exercise. The task force that build the Maturity Model leveraged just about every task force and group within The Green Grid’s Technical Committee to come to a consensus on what we felt the first pass at this was, and we also felt that even though we had done a great job on this. That this was not going to be the final say. That this was only Version 1 and it was going to take refinement and further versions to really get to the final product.
But, we came out with this and, you know, when you – the content here, the potential guidance, the types of targets that are available are – really show an astounding level of both breadth and depth. Both breadth across the data center and depth in terms of specific targets and specific goals you might want to achieve.
The way we foresee organizations using this is really to do a quick assessment of themselves now, kind of score where they are on these different sections and subsections, and then plan out five years from now, “Where do we think we would like to be? What do we think would be appropriate targets for us?” And then, take a look at the gap between where you are today and where you want to be five years from now, and start thinking about what your process is going to be to manage that transition from today to 2016.
We’re very excited about this. The reception that we received upon the introduction of this has been just about unanimous in a claim people have been very excited. We have had a number of organizations that have said, “You know what, we’re going to help you pilot this. We’re going to go test it out. We’re going to tell you where the bugs are and we’re going to give you information that you need in order to do Version 2.”
And this is something that, as I mentioned before, it’s not a one shot deal. We’re looking forward to having this be a major part of The Green Grid’s guidance going forward for a long time to come.
Mark, I think unless you’d care to add anything about the Maturity Model, we can probably pass it back over to Scot for Q&A. So Scot, why don’t you go ahead and take over and bring us into questions-and-answers.
Scot Heath: Sounds good. We do have a few questions, John.
Rebecca Franco: Okay, the first one, “What level of availability data center Tier 1 to 2 – oh, I’m sorry, 1 to 4 will be included in the simple definition of PUE?”
Scot Heath: Oh, well, good question. You know, PUE is separate from – you’re talking about the Uptime Institute’s definition of Tier 1 through Tier 4, which are really meant to target the maintainability of a data center, how I maintain the internal systems.
And PUE, on the other hand, is a completely separate definition, you know, meant to target specifically energy use efficiency, and of course the definition of that is total energy into the building divided by the IT component of that. So, we purposely don’t put, you know, a tie into the tier levels in there.
Now, we also recognize that Tier Levels, or maybe not necessarily strictly Tier Levels, but the structure of your data center will drastically affect the PUE that your individual data center is able to achieve, and this is a big reason why we discourage comparison against, you know, the data center across the street.
So, the big guys out there, and you’ve all seen them publish, you know, numbers, you know, best PUE achieved, blah, blah, blah, low PUE achieved, and we encourage that. I mean, I’m all for having people drive these numbers as low as possible but to say that, “I’m not doing a good job because I can’t attain that PUE,” is not necessarily true.
So, I guess the bottom line is that, you know, while we don’t implicitly have that definition embedded in the PUE, we recognize that it’s important.
Do we have another…
Rebecca Franco: Great.
Scot Heath: …question Rebecca?
Rebecca Franco: Yes.
Rebecca Franco: “You didn’t mention partial PUE. Can you talk in more detail about that? What is the difference between PUE and partial PUE?”
Scot Heath: Sorry, that was definitely my fault. I had that on my speaker notes to do and I blew right by it, so let’s do spend a little time on that.
So, partial PUE was developed in response to people’s notion of how they carve up a data center to measure power. And sometimes, you know, that got interpreted quite liberally, and so you know it is valid to want to compare different technologies used in the data center.
For instance, I may have (slides) where the UPS is (unintelligible) and I may have battery UPS’s, and I may have, you know, native 208 and native 480 UPS’s, and I may want to know, you know, how all those different things affect PUE in a standalone fashion.
And so, partial PUE gives me the ability to draw a box around a portion of my data center infrastructure that includes IT equipment. I have to have that because, you know, the denominator in my fraction is power going to the IT equipment and if I don’t have any in there that’s 0 and my result is undefined. So, I can draw the box literally around any piece of the infrastructure I would like and define that to be a partial PUE measurement.
Now, that measurement is useful for me as an operator perhaps to compare those different technologies. So, I could have different halls, I could add different buildings, I could have different whatevers; different sizes. And to calculate the ratio of just my UPS technology, maybe the PDUs are included in that and the IT equipment, you know, that I’ve drawn in my box, against one another, is a perfectly valid thing for me to do.
It’s even more difficult with partial PUE to draw conclusions against, you know, someone else, depending on how they, you know, happen to have defined that, and even if they choose to publish that. But, it’s a document that, you know, a guideline defined way to go help you compare different technologies in the data centers.
Power deliver is one example, cooling technology might be another example, I may want to exclude, you know, the UPS component from a partial PUE discussion and just look at different cooling technologies. I just bought, you know, some new wiz bang whatevers and – with a claim, you know, increase in performance over my existing ones. I can go measure that now and find out how that looks by drawing the box around those particular pieces of equipment.
Okay, what’s next, Rebecca?
Rebecca Franco: Let’s see, “Can you talk more about CUE and how it is measured?”
Scot Heath: So, John or Mark, would you like to take a whack at that one?
Mark Monroe: Sure. Carbon usage effectiveness is the – you can think of it as the carbon intensity of your IT equipment. One interesting thing I think that John mentioned was that the denominator of these fractions are – or of these ratios are always the same, so we tried to make it as easy as possible to collect that information. So, all you need is once you’ve collected the IT power usage then you can use it for any of the fractions.
With CUE the numerator of that fraction is the amount of carbon generated by the generation supply, the generation equipment that supplies your electricity in the – used by the entire facility. And so one way of collecting that – the simplest way to calculate CUE, if you break it down into what’s called the Carbon Efficiency Factor, which you can collect from usually data produced by federal or international energy agencies, and in essence it’s the amount of carbon that’s produced per kilowatt hour of electricity that’s used – that’s produced in a data center or in a electrical plant.
So for example, the statewide average for California, I believe, is 0.8 pounds per kilowatt hour of electricity, and in Colorado where I’m located and Scot’s located, it’s somewhere around 2.2 pounds per kilowatt hour of electricity produced. You multiply that by your PUE and to – you end up with carbon usage effectiveness.
So, the carbon emissions factor and – times PUE gives you carbon usage effectiveness, or you can take directly the amount of carbon that’s produced and divide it by the IT energy.
So, I hope that makes sense.
Scot Heath: And I think we should probably note, both of the CUE and WUE, which we talked about, are not unitless elements…
Mark Monroe: That’s right.
Scot Heath: …anymore like the PUE is. PUE is power and power and energy and energy. These actually have got gallons associated with them or pounds or tons, or whatever…
Mark Monroe: Right.
Scot Heath: …you measure the carbon in.
Mark Monroe: And it’s important that the denominator of those be kilowatt hours then as well so that the – it’s an energy usage on the bottom of the CUE and WUE factors.
Scot Heath: Always, always, always.
Rebecca Franco: Okay, we have another one here. “Has there been any consideration to incorporating the possible DOE and EPA future requirements for carbon footprint reporting, or is this seen as something totally different?”
John Pflueger: I’ll take this one. I think we’ve been looking into it and, you know, kind of going back to the model that Mark presented at the beginning of the webinar on the lifecycle of a metric, we’re at the beginnings of CUE and WUE. And you know I suspect that there are going to be some changes there.
I also know just from other areas of the business in which I work at Dell that there’s more and more interest in general in the IT industry around carbon reporting and emissions reporting. So, I don’t exactly know how it’s going to play out yet, but I think there’s going to have to be some sort of convergence between the two sets of requirements in the future.
But, it’s – I don’t think it’s built in there as of yet. Mark, would you agree?
Mark Monroe: Yes, I’d agree. And I think you see from The Green Grid’s past efforts in harmonization that might be something that The Green Grid would be particularly good at. We’d be bringing together those different views on carbon reporting.
John Pflueger: I’d also say that, you know, there are steps through the kind of metric maturity model that we can progress through while we’re waiting and kind of while some of these requirements are maturing.
So, while the requirements are maturing and while we’re trying to understand what reporting is going to mean to the data center, you know, we can start the process of figuring out, how do we collect meaningful data, how do we do it in a repeatable way, and how do we do it in the – in a – with a rigorous protocol that everybody can look at and go, “Yes, you know, that makes sense to us.”
Rebecca Franco: Great. Thank you. And speaking of the Maturity Model, the next question is, “Can we get a readable version of the Data Center Maturity Model chart, or is this for Green Grid members only?”
John Pflueger: Mark, you’ll have to verify, but I believe the full Maturity Model is on the public site of The Green Grid’s Web site.
Mark Monroe: Yes, I think so too. The easiest way to tell is to hop over to The Green Grid Web site and look under the Libraries and Tools section and you’ll see the Maturity Model will come up. And if there’s a little lock next to it – actually I think the Maturity Model itself is open content, but if there’s a lock next to it then it’s members only.
Scot Heath: It actually is open, Mark. I just logged…
Mark Monroe: It’s open?
Scot Heath: …out and took a look at that and I’m able to pull that chart up. I can make it big enough that I can actually read it.
Mark Monroe: Great. Yes, that’s – yes, the intent is…
John Pflueger: And…
Mark Monroe: …that, you know, you pull the PDF down and print it as a wall-sized plot, and then start using it as a…
John Pflueger: Yes.
Mark Monroe: …visual tool as much as anything else.
John Pflueger: And, you know, if – and by the way, just in my own check, I think you’ll – I think folks in the audience will find if they use their own particular favorite search engine and just type in the works Data Center Maturity Model, they can get right to it.
Mark Monroe: Great.
Rebecca Franco: Great. Thank you. Let’s see, “The Green Grid just released PUE’s 0 to 4, can you talk more about the differences between the four types?”
Scot Heath: Well, I can cover that in maybe a bit more detail. You know, as I mentioned previously, the 0 to 4 are meant to be increasing levels of accuracy of PUE, and really what we’re doing, you know, in the last three, 1, 2, and 3, are getting closer and closer to the IT equipment itself so that we isolate more of the energy used in the data center as losses, versus you know, used to power the IT equipment.
So, depending on, you know, a power delivery methodology, if you have 208 volt UPS, then you won’t have a PDU, and so you may not be able to do a Level 2 where you’re measuring past the PDU, so in that case those kind of numbers collapse.
Best of all, you know, currently in the definition is to measure the power at the IT equipment and if the IT equipment has power reporting capability, you can certainly use that to measure that. Many people have, you know, brand circuit monitoring at the power strip level, either on a per plug or on a per power strip basis. That would be considered to be at the IT equipment – input to the IT equipment.
Level 0 differs from the other three levels in that it’s a demand-based measurement. That is, I go look at watts, I don’t look at watt hours consumed, and the danger there is of course that watts continually change during the day. My IT equipment takes different watts, my cooling equipment takes different watts, and depending on the frequency of my reading, I limit the accuracy of my, you know, PUE calculation because I’m not really looking at the total energy consumed, and that’s really what we’re interested in for the definition of PUE.
So, if I have some way of recording that with a watt hour meter or I sample often enough that I can integrate, you know, between samples, then I can actually record total power or total energy used, and then I’d fall into the 1, 2, or 3 category.
Again, you can certainly find the guidelines for this on The Green Grid Web site. It’s very straightforward…
John Pflueger: Yes.
Scot Heath: …and I say that tongue in cheek. As I said before, we’ve had many, many discussions about, “Where do draw the boundary?” You know, things as simple as…
John Pflueger: Yes.
Scot Heath: …I’ll have a conference room and it’s used by people other than data center people. Do I get to exclude it? You know, it’s got fans and an ice machine and whatever in it, so it’s not always as simple as it would appear on the surface.
Mark Monroe: You know, Scot, and I’ll just add that in some cases where members of The Green Grid or others have measured PUE over time, done instantaneous measurements in a series of those every 15 minutes or every 5 minutes.
We’ve seen variations within a single data center of as much as 15% or 20% between the high and the low PUE measurements based on the IT load that’s going on at that particular moment, what the weather is like outside, the load factors of the various equipment that’s on there.
So, that’s what – why we’ve gone to the – the higher accuracy requires collection – sampling over time and averaging, because you – we really want to give a better impression than you know, “I took it at 8:00 am on a Tuesday afternoon – or Tuesday morning when the temperature outside was low and all my economizers were running, and what do you know, I got a 1.1, but my average is somewhere around 1.3 or 1.5.”
Scot Heath: All right. Well, I think that we’re out of questions here. I really want to thank Mark and John for joining us today.
Rebecca, do you have any closing remarks?
Rebecca Franco: Yes. I wanted to remind our participants that the replay will be available on our Web site with 24 hours, and that’s www.42U.com. If you feel like your questions were not answered today, I would like to invite you to call us at 1-800-638-2638 or send your questions via our Project Evaluation Form on our Web site.
Have a great day.