Exploring Key Aspects of Data Centre Design

[00:00:00] Speaker 1: Good evening, everyone, and welcome to SIPC Victoria's webinar tonight, exploring key aspects of data center design with Chris Wallace from Mace. Sorry. SIPC would like to acknowledge the traditional custodians of the land, the Boon Oong Boon Wurrung and the Wurundjeri Woi Wurrung peoples of the Eastern Kulin Nation and pays respect to their elders past, present and emerging, and extend that respect to any Aboriginal or Torres Strait Islander people here today. Tonight, we're just going to go through some SIPC announcements and then shortly Chris Wallace from Mace will be presenting, followed by a Q&A at the end. And I strongly encourage everyone to put in your questions in the Q&A function just in the top of the screen, and then we'll go through those at the end of Chris's presentation. The 2023 Young Engineers Awards is open for entries to undergraduates, building services graduates and emerging professionals. The theme of the 2023 competition is centered around embodied carbon, an issue that reaches into all areas of building services engineering. The awards call for the skills and knowledge to better educate the industry and address the challenges of reducing embodied carbon in building services. Entries close 1st of August 2023. You can visit sipci.org.au for more information on entry criteria and entry inspiration, including tips from our judges. For those who no longer classify as emerging professional, nominating a mentee for one of the three categories is a great boost for your organisation's developing talent. [00:01:57] Speaker 2: The next slide is a great boost for the future. [00:01:58] Speaker ?: The next slide is a great boost for the future. The next slide is a great boost for the future. The next slide is a great boost for the future. [00:02:01] Speaker 2: The next slide is a great boost for the future. The next slide is a great boost for the future. [00:02:03] Speaker 1: The next slide is a great boost for the future. The next slide is a great boost for the future. The next slide is a great boost for the future. The next slide is a great boost for the future. The next slide is a great boost for the future. The next slide is a great boost for the future. The next slide is a great boost for the future. The next slide is a great boost for the future. The next slide is a great boost for the future. The next slide is a great boost for the future. The next slide is a great boost for the future. The next slide is a great boost for the future. The next slide is a great boost for the future. The next slide is a great boost for the future. The next slide is a great boost for the future. The next slide is a great boost for the future. The next slide is a great boost for the future. The next slide is a great boost for the future. The next slide is a great boost for the future. The next slide is a great boost for the future. The next slide is a great boost for the future. The next slide is a great boost for the future. The next slide is a great boost for the future. The next slide is a great boost for the future. The next slide is a great boost for the future. The next slide is a great boost for the future. The next slide is a great boost for the future. The next slide is a great boost for the future. The next slide is a great boost for the future. The next slide is a great boost for the future. The next slide is a great boost for the future. The next slide is a great boost for the future. The next slide is a great boost for the future. The next slide is a great boost for the future. The next slide is a great boost for the future. The next slide is a great boost for the future. The next slide is a great boost for the future. The next slide is a great boost for the future. The next slide is a great boost for the future. The next slide is a great boost for the future. The next slide is a great boost for the future. The next slide is a great boost for the future. [00:02:46] Speaker 2: The next slide is a great boost for the future. The next slide is a great boost for the future. [00:02:48] Speaker ?: The next slide is a great boost for the future. The next slide is a great boost for the future. [00:02:50] Speaker 2: The next slide is a great boost for the future. [00:02:51] Speaker ?: The next slide is a great boost for the future. [00:02:52] Speaker 2: Don't miss a SIBSI event. [00:02:54] Speaker 1: Sign up to receive the latest industry news and events from SIBSI ANZ region. Register for your SIBSI regional e-newsletter at sibsi.org.au. SIBSI would like to thank Connexus Recruitment for supporting SIBSI ANZ's annual program of technical seminars and young engineers networking events. Connexus Recruitment are a boutique recruitment consultancy dedicated to the building services design and sustainability sectors. For over a decade, they have cultivated extensive networks and a strong presence throughout Australia and New Zealand's design and construction sector, enabling them to connect some of Australia's most recognised design consultancies, contractors and client side entities with talented individuals from across the globe. [00:03:44] Speaker 2: Now for what we've all been waiting for. [00:03:47] Speaker 1: Chris Wallace is an expert at managing complex engineering and IT projects where reliability and availability is essential. He has a wide experience of data centres, high rise buildings, fit outs, green and brownfield sites, and overseeing the APAC regional data centre business unit, which is responsible for more than 100 megawatts of hyperscale data centres. As head of data centres and critical facilities for MACE, Chris's insight is invaluable in this rapidly evolving space. And now I will hand over to you, Chris. [00:04:20] Speaker 3: Thanks Matthew. I'll just show my screen up on me. Can everyone see the PowerPoint? I'll just put it in present mode. OK, thanks everybody. You can see my screen, yeah, Matt? Yeah. [00:04:38] Speaker ?: Yeah. [00:04:39] Speaker 3: Yeah. [00:04:40] Speaker 4: Great. Thanks everyone for enjoying me. One of my colleagues in Mason Australia has kindly volunteered me for this and he's last minute not on the call. So I don't know whether that's a good thing or a bad thing, but bear with me. What I want to do is just talk a little bit about data centres in general. I'm going to do a little bit of an introduction, then walk through electrical, mechanical, fire, IT, security, commissioning, and then close, just briefly close, which is really the constituent parts of data centres. One thing I find about data centres is there's lots of things that are interesting and complicated, but one of the things that's more than ever, the customers are very, very aware of intellectual property. So the whole of the presentation this evening cannot be in very much detail and it cannot be designated in any shape or form to any one of our customers or clients or anywhere like that. So it's a bit generic. I mean, hopefully it's not too generic to be, you know, a waste of time or of no interest. But, you know, bear in mind that if it's something that's leading edge that, you know, one of our customers is doing, you know, we've got a non-disclosure agreement. So we can't actually tell you all about it, much of though we might like to. The other thing is, is I don't really know the audience that well. I don't know whether people are very expert in data centres or never been to one. So I've tried to hit some sort of middle ground. So again, please bear with me if it's either, you know, massively too simple for you or on the other hand, if it's something that there's some element of it you don't understand, which probably unlikely. Anyway, so I'll head over and before I do talk about, you know, the data centre, I have to give the obligatory introduction to the company and MACE. MACE basically are originally UK based, but now global construction advisory service. They do project management, cost management. In actual Europe, they will actually be your contractor and build things for you. Throughout Asia, we're we've got quite a sort of widespread and we are offices in the usual, you know, hotspots like India, Macau, Hong Kong, Philippines, Vietnam, Singapore. And in Australia, we're in Sydney and Melbourne. Not really much in ANZ in New Zealand, but we have got a presence there. Tonight I'm speaking from Malaysia and that's why Malaysia on the box has got a little bit of a different colour. As I say, MACE really have been in Asia for a long time, never really had hit the data centre scene in any shape. Big size until I was brought in just less than a year ago. And when I was brought in, we subsequently won two projects in the Philippines, which are underway at the moment. And we're very proud of those and we're looking at some more proposals even as I speak for more projects in the Philippines and elsewhere. Elsewhere, you know, for some of the key customers, our key customers in Europe are largely Microsoft Meta. And there we will actually, as I say, build large hyperscale data centres for them. Personally, I've been in Asia since 1997 and really did, you know, my background, obviously, I'm simply a member, a professional engineer. My background has really been in the sort of project management side of the business rather than the design side of the business. So this is really by way of observations of the design, as I've seen it going through, rather than necessarily a recipe book as to how, you know, I would sit down and do a design if I was doing design. And I say the first data centre per se that I did was way back in 1999, where we were converting to try and get speed to market, converting a building, a factory building in Singapore into a data centre. So I've been involved in and around data centres for quite some time, probably 20 odd years now. So I thought it might be worth just starting off with this to talk about what, you know, what are data centres? And I think the top right picture there is a sort of, you know, classical view of what people might think of as a data centre. And that's probably a slightly older model of data centre. But essentially, really, the buildings which are obviously interconnected to each other, if they're standing by themselves and not talking to the rest of the world, they're not really much good. And they house rows and rows of a cabinet, you know, computers, servers, storage, and of course routers and switches and everything that makes the whole thing work together. So these are that the picture on the top right there shows typically what these cabinets look like. And in terms of getting an idea as to, you know, how how many you get, on average, you probably get about one per three square metres of the building white space area. And I say that the white space area is the area that typically used to have a raised floor in it. And the raised floor used to be typically white, high pressure laminate, so that they came up with the term white space. And, you know, one to three square metres is OK. You can do a little bit better, but one is one to three is a good jump figure, a rule of thumb. And we'll come back to that later because it's one of these things where you can actually use that sort of building block to actually do quick rough guesstimates on things, you know, very, very quickly. Just saying, well, if I'm thinking about one per three square metres, you know, then I can go from there. Data halls, really, the white space, they're not all the story by any stretch of the imagination. At the back of that, you've got all the cooling and the switch room. I say cooling in this part of the world. Typically, you know, we still are refrigerating or cooling or one way or the other data centre. Elsewhere in the world, obviously in the Nordics and, you know, some places in America, they can just blow air through, temperate air through, and that will be sufficient to cool the building. But, you know, typically in Asia, find that you need, you know, the white space won't be the whole building. It might actually be between 40 to 60 percent of the overall built space. You might get a little bit more than that, but probably not much more than 60 percent. If you get 60 percent, you're probably doing OK. And then obviously, you know, we all know or anyone that's sort of been in around data centres, it realises that the reliability of the data centre is a key concern. And this is basically, I don't know whether people have come across it, but there's an application you can get now called Down Detector. And if you go on Down Detector, it will tell you whether your favourite website is actually working or not working. And it will tell you that, you know, it had a crash for this reason or that reason from this time to that time. So the customers are really, really, you know, either banking or you're just cloud computing or your travel agents, whatever. Yeah, they're really they don't want that their app to be down. So they're very keen on reliability. So this really starts from the kickoff. When you're going to look for a site, you want to find a site that's unlikely to cause you a problem. So the typical things are it's away from floods, it's away from earthquake zones. Obviously, you know, some places in the world you can't get away from earthquake zones. You know, the Philippines, Taiwan, many other places, Indonesia, you know, you've just got a design round it. But in an ideal world, you'd like to be away from earthquake zones. You want maybe you want to be away from sources of electromagnetic interference like, you know, overhead railway tracks or highways or, you know, in some cases, airports. There's there's a whole rate list of things that are sort of things to look out for. And one of the metrics when we look at how reliable a data center is or that there's two basic metrics that the industry use. One is the uptime Institute uptime being the opposite of downtime. And the other is the EIA TIA 942. So uptime Institute a while ago was started off as really like a boys club of data center owners and operators. And it was sharing, you know, freely information as to what they could do to make sure that their data center was up and not down. And they came up, you know, quite quickly to some concept they called tears. And then eventually they commercialized it and monetized it. And now the word tears actually is actually trademarked. And so when you look at EIA TIA, who have done a very, very similar thing, they come in early days, cooperated with uptime Institute. They can't use the term tears anymore. They've got to call it rated. So basically in simple, I'll use the word tears for one of a better but but rated is almost the same. So tier one is very, very little, you know, resilience and, you know, very basic bottom. Very, very simple. Very few people I've ever come across that have built a tier one data center. Tier two is slightly more. It has a little bit of resilience. Tier three is typically what people will buy and most of the market is going for or rated three in that if they're using TIA 942. And, you know, tier four is, you know, a much higher level of resilience. So basically the difference between tier three and tier four and, you know, you could write a whole lecture on this alone is tier three is what they call concurrently maintainable. And concurrently maintainable means that you have typically for major capital plant items rather than terminal equipment n plus one units where n is the number that you need to do the job. And you have another one, a spare one. And the idea is if you have a generator or a chiller or a UPS unit or whatever the unit is, if you need to actually work on one of them, you can turn one off and you still have n left to work and the building doesn't go down. Whereas fault tolerant means that you have the same end plus one situation as a minimum because it's actually concurrently maintainable and fault tolerant means that whilst you're in that maintenance situation, if there's a problem, you don't lose the services. So that means you have two end, which is basically means for all the major capital plant in the building, you have to, you know, if you need one, you have to, if you need four, you have eight, basically two end systems are almost duplicate systems. It's almost like having two full sets of buildings because of the cost of that, then not often used. It might be for something like an air traffic control center. Like that or a government, you know, you know, high security, you know, GCHQ type of environment. But for most of the banks and most of the cloud computing people, most of our customers, they won't want a tier three. Or rated three. Just another point on the Uptime Institute and EIA, EIA. Until recently, Uptime, as I say, they very quickly monetized the whole concept where it started off as just, you know, industry colleagues helping each other out. Then it got the stage where they'd monetize it. So they would then say, well, we can train you to become a tier specialist and a tier qualified tier designer. When you, you know, to do that, you know, anyone who's going to design a tier compliance scheme must use a tier trained person. You know, and then you have to have an audit done by qualified tier certified auditors. So it became very monetized. Whereas EIA TIA 942 was really a standard in the US and it wasn't monetized so much. However, they've just started to actually restrict the people who are now doing audit. So you have to be qualified to do it. There's a list of approved auditors now. So there is that they're already starting to get into that sort of monetized sort of situation before, you know, anyone could go and do it. Any IAT IA 942 order. Now that's not the case. The other big issue about data centers, and this is the one that I guess is probably at the heart of why they become a little bit controversial, shall we say, is the increasing power consumption. And the power consumption is often quoted as a power density, and that's either in watts per square meter or watts per rack. I didn't before, you know, it used to be about three kilowatts a rack, you know, that was not a bad density that you would see. Now, eight to 15 kilowatts per rack is common. It's not that unusual to say up to 50 kilowatts per rack. So in basic in terms of watts per square meter, you know, it used to be about a thousand watts a square meter. It can be 5000 or even 16000 watts per square meter. You have huge, huge power loads on these data centers. And it's because of these power loads that people like the Singapore government recognizing that the data center is now data centers are now probably consuming 25% of the power in Singapore. But they're saying, you know, right, they're not providing that much employment. They're creating a lot of waste heat. A lot of them in Singapore are using a lot of water as well. So they put a moratorium on building new data centers in Singapore unless there's some real key defining attribute that makes it different. It's got some whiz bang green or low energy background to it that may make an exception. But for the average person, you're not going to build a new data center in Singapore very easy at the moment. There was a guy. I think it's Christian Baladi. I think his name is from Microsoft. He coined the term power utilization efficiency of PUE. And this was really to try and make the data center business a bit more efficient. And this was really a ratio of the total power used in the data center divided by the power useful for just driving the computers. So basically, if you have a data center that is a 10 megawatt data center and that will mean a 10 megawatt of data power, but it's actually using, say, 15 megawatts of power at the incoming supply from the grid. That would have a 1.5 PUE and typically the PUE of 1.5, you know, most people would not be satisfied with most people would only be satisfied with a 1.4 or below as a PUE. Why I say that is a lot of people were trying to get you down below 1.3, maybe into a 1.2 by using some, you know, specialized chillers or something like that. But in many areas that we're dealing with now, water supply is not necessarily readily available. And they're going to go for a air cooled chiller. And once you go for an air cooled chiller, the efficiency drops and obviously, you know, the power consumption goes up and you may be up at a 1.4. But that 1.4 really is about as bad a PUE as most people would expect. You know, if you can't do better than 1.4, you need to have a look at the design again, I would suggest. But anyway, so I've given you a little bit of a background in terms of the general thing and talk a little bit about electrical. Before I flip on the next slide, I'll say that the electrical cost can be 40% of the project cost. So, you know, you've got a building that might cost you 200-300 million US dollars. 40% of that is probably going to be electrical. So, you know, how do you put together a scheme with high loads and high reliability? One of the things that's nearly always become a requirement is to have two separate grid feeds from the national grid or whatever local grid is from two different substations. Now, it's not always possible to do this, but this is what most people would want if they can achieve it locally. I mean, I remember years ago working on trading floors in the UK and, you know, LEV London Electricity Board would not let you have two separate power feeds in the building because they wanted to kill one power feed if there was any danger or fire and then know that the power was off in the building. But here exactly the opposite is true. They don't want to use the power in the building, so they want two supplies. Often because of the size and to try and keep the conductor sizes down, they will try and distribute around the building at medium voltage. So, depending, you know, on the size that, you know, is a typical rule of thumb. Most data centers are below above 30 megawatts would or 30 MVA because, again, with a with a power load, a data center power load, the power factor is almost always unity or close to unity. So for megawatts and MVA become almost the same when you're talking about the data side load. When you talk about, you know, anything that's getting above about 30 MVA, you're probably going to go to a high voltage power supply. And depending on where you are in the region, you know, it would be different voltages. So I won't quote voltages in different countries, different things, but quite often what they do is is come in at high voltage and then drop down through transformers. You obviously have a supply, local supply authority, high voltage switchboard, you might have an owner's high voltage switchboard as well, depending on the situation. Then you drop down through a transformer to owner's medium voltage switch gear. And then what I found over the years is what's become more and more common is you will run things through a series of ring main units. So you run rings of ring main units, loops of ring main units. So and what you'll do is you'll have the ring main unit loop, ring main unit loop fed from the two different feeds, but we'll have an open section in it. So where the open section is depends on the design concept. It may be that it's open in the middle of the ring. It may be that it's basically open at the far end of the ring and that you will normally run on one supply and you will only use the other, you know, isolate the other half of the ring and open and close the ring from the other side in the event of a failure. But that's typically what you'll do. So ring main units for anyone who doesn't know is normally a main, a medium voltage thing where you have to you have an incoming supply and an outgoing supply, which allows you to form the ring. And then between the incoming and the outgoing supply, you have a tap off, which allows you to feed the load. I'll show you in the next slide what you have. So typically what you do is you set these things up and each ring main unit feeds a block and the block again show you on the next slide will normally be determined between by the generator size, because what you're going to do is you're going to say in the event of my mains failing and by losing the, you know, all the power from both sides of the ring. I'm going to need to feed that block from a generator and therefore the generator size and the generator is going to determine typically what that block is. How many of the blocks you use will depend on what the, what the load is in your data center. So it may be that you consider this as you have, you know, one of these would represent a ring main unit represent maybe one data hole and the data holes may be, maybe on, you might need, depending on the size and the color. You know, capacity data hole, you might have two on one floor or you might have one on one floor or you might have more than one on one floor. So really these are just, yeah, this is the sort of the building blocks. And as I said before, this is, you know, not what one particular company does. This is what I found over, you know, with variations over, you know, 10 last 10 years, most of the customers that we've been doing have been doing something like this, you know. So this will give you basically, if you look at it as the red and blue as an N plus one, and the N will depend on, you know, how many of these ring main units or how many blocks you need to feed your load. And so what I've just said in brackets there is the N plus one is really only for, you know, major plant. For terminal equipment, you may have N plus 25%. So if, for example, you're talking about cooling units, you would not go for N plus one because that wouldn't give you, yeah, the units would be too big to actually work as a unit. So like that, obviously, if you go to N, you're rated, this is what I say, you're rated for tier four or rated four, that suddenly gets very, very expensive. And this is just generally the scheme that you'd use to support mechanical or, or electrical works. But really how much and how well you support the mechanical systems really normally is a function of the chiller size. Nearly always the blocks that you're using and the chillers that you're using are really going to be a bit big to start on a UPS unit, or the UPS is going to be quite a bit, quite a sizable UPS to actually, to restart a chiller. And the other thing is the chillers typically, unless you've got a chiller with magnetic bearings or some other modern scroll chiller, it's going to have a, you know, a fairly substantial restart time. So, you know, the fact that you, you know, if you're not going to keep it on a UPS system, you're going to have to have a pump it down and rebuild it back up. And therefore, you're going to, what most people do is they end up having some form of thermal energy storage, which means basically they make and, and support, you know, cool water. They have a tank normally or tanks with, with a store block of cool water in, and that will allow the chillers to run for, sorry, that will allow the pumps because you can support pumps fairly easily on a UPS. That will allow the pumps to run for, you know, maybe 10 minutes, and that will give you a chance for the generators to start the, the chillers to shake themselves down and get themselves back in working order and things like that. But anyway, going back onto the next slide on electrical, what I described on the last slide of the ring main units, you can see those at the bottom of each of these blocks here. And you can see, you know, one in one out and one spur. And so again, I've used the, the red and the blue is to come to show you that they, they may be from opposite ends of the ring. And what you then do is you come from the MV ring main unit into the transformer. And look, bear in mind, these are actually just very, very dimatic diagrammatic. You know, if you wanted to go into all the details of how the data centers do these and, you know, how the fact that quite often all of this is automated and not only is automated, it's also metered and you have a low voltage system, which is actually looking at all of this. And you can actually remotely operate from a keyboard, you know, open and close breakers from the keyboard. You know, that that's how it works. But this is really just for simplicity to trying to keep it to a relatively simple and relatively short presentation. So basically in and out at MV, you know, with a feed off to a transformer after the transformer into a low voltage switchboard and then out of the low voltage switchboard, you have an ATS and automatic transfer switch to a generator. So obviously those two incoming breakers are locked. So you'll either have the generator or the feed from the transformer, but not both. And then you'll have a feed out to a UPS to do the IT load and then maybe a non UPS supported load, which might be a mechanical system of some sort of a pump or something like that, or a fan or something. But it's not something that you're actually mission critical. So again, what you typically do is you'd have the, you know, the red and the blue feed and then what you have there at the bottom is another block and that block would be a redundant block. So basically you'd say, right, I've got N might be two and the plus one is the redundant block here and that redundant block might be one redundant block per whole or it might be one redundant block which will serve all the red and blue units. And yeah, and this is the difference between basically block redundancy or distributed redundancy or in some cases you might go for a combination of both. You know, you know, the blocks are typically two and a half, three MVA generator size. The generators would normally again depends on what you want and where you are and whether you're following just, you know, some sort of a specification from, you know, you know, the client or whether, you know, some. Some other paper that you're trying to follow, like, yeah, uptime or or TIA, you know, the typically the fuel storage is underground tanks between 24, 24 and 48 hours, sometimes up to 72 hours. But that's when you think of, you know, if you've got a large data center and you've got a lot of generators on there and you're running, you know, tanks underground tanks for, you know, fuel for a lot of big generators. You know, you end up with a huge tank farm and a huge supply of oil sitting there and then you end up having potentially having problems with the oil sitting too long, but between usage and so on and you end up with fuel scrubbing and things like that. But no, this is just a general thing. So basically just the load path from the MV through the LV or from the ATS if the if the power is off to the UPS to the data cabinet and a similar block might be for mechanical without any UPS. It might just be or maybe only a UPS for the pumps and the fans, but not the chillers. And occasionally we see some of the customers where they don't really want a big centralized UPS. What they will do is they'll say, right, all my servers will have their own inbuilt UPS with them. So their server will be almost like a laptop, so it will have its own little battery pack with them. Some use this as a little bit of a cheat because what they do there is obviously the UPS has some losses. It's normally quite a high efficient thing, but it has some losses. And if you worked out of saying the actual PUE of 1.3 or 1.4, some of that PUE is UPS loss. But if you're actually supplying a server which has got its own built in UPS, you might be a little bit cheeky and say, well, that is actually only supplying data, but there is actually a little sort of tricky bit of UPS loss in the onboard UPS system if that is the case. The other thing is typically for the generators, there will be a separate feed out to typically a common buzz bar where you will have a load bank on the roof and you might have a feed out from each of these LV boards as well. So the idea is that you can actually look and test out the generators every month without taking the data center down and that's what most of the customers end up doing. And because basically a lot of this is done in buzz bar rather than cable, you know, buzz bar will become a significant expense as part of the overall project value. They say a picture is worth a thousand words. So typically what we've got here is in the cabinet at the bottom here. Typically you'll have a power strip on either side of the cabinet and the power strip will one will be red and one will be blue. You know, as we said from each of the blocks and each of the server units or, you know, bits of IT equipment that the customer sees what they call dual corded. And that means it has two cords and it will feed from red or from blue or red and blue. And what it will do is it will sense the power all the time and it will have a little static transfer switch in it. And what it will do is it will say, OK, blue's gone down. I'll just change to red. And so basically if you lose one supply completely, all the cabinets set up there because they they're all working with these little bits of equipment. They've got two cords on them. One of the issues with this is is you just a little bit on the top right there. And that's sort of gives you a little bit of a view of, you know, a simplified schematic, which I thought was quite nice. So you basically go from your your MV through your HV through your LV and then into the data halls with the red and blue power strip down on top of the cabinets. And then you come off the red and blue power strips in each cabinet down to the power strips in the cabinet like that. But one of the things that's, you know, people try and get more and more clever with these data centers. So they ring the changes instead of having maybe n plus one or, you know, situation. They might come up with a scheme that they call three n two or four n three, where that means that, you know, n is the number that is three and any two working will be enough. Or four is the number that you need of which three need to be working. So the little three colored UPS block on the on the right hand side there is actually a three n two scheme. And it's showing the first server, which is taking a feed from blue B and C. And the second server is taking from A and B. And the third one is taking it from A and C. So one of the things that, you know, you've got to be very, very careful. And there's been outages in buildings caused by this is when you actually load up the servers in each of these cabinets or in each of these rows. How much are you got on each thing? Because if you haven't got the same amount total on A as you have on B, as you have on C, when you actually switch over from one to the other, you can actually overload one. So what can be a real problem is you come up with a scheme which is actually inherently complicated and the poor souls that end up running it have got to keep an eye on everything that switched. You know, basically when they rack something up, they say, right, OK, I'm adding so many KW to A and B. OK, well, when was the last time that does that throw that over the balance now because I haven't added another kilowatt to B and C or, you know, et cetera, et cetera. So that's that. And then the other just final thing to before we go off this is that in some cases these power strips are well, they're nearly always monitored to work out how many, how much, how much their actually power has been delivered there quite often to bill if it's a customer that's, you know, selling the space on to bill it to their customers. But if it's not to bill it just to keep an eye on, you know, the balance of it. And then in some cases you can have the powers, the sockets on these power strips like in that cabinet there can be rebooted remotely. So you can have a situation where if you've got a server in there and the server has gone down and, you know, the old trick, you know, control or delete or hit the on off button, the God switch. You can do that, you know, from your office in Miami when the data centers in, you know, Singapore or, you know, Shanghai, you know, if you've got all the right permissions and everything else like that. So what you find is there's a whole subset and this is, you know, really MVL, touched on HV, HV, MVLV. But there's a whole subset of ELV systems and controls like the power management system, which get incredibly complicated, incredibly, incredibly quickly. So just moving on a little bit more. When I was talking through the slides with Sean, my colleague, he said, what about UPS? And he said, the static versus rotary, you know, don't go there. You know, static versus rotary UPS is it could be a complete separate presentation. In short, the top two boxes are a very simple, you know, representation of a static UPS. Basically, you have the power coming through. It's not quite normal. It's not quite normally how it happens is that, you know, it's showing basically going through as raw mains and then feeding a battery charger and a battery. What would normally happen is it would actually go through the battery charger. The green line should go through the battery charger and the inverter to the load. It wouldn't normally go through that bypass line. And simply what happens is you charge it basically turns the or you have an inverter which turns the AC to DC. Then you have a rectifier change the AC to DC. Then you have an inverter which inverts it back from DC into AC. And in the event that the power goes off the light, you know, someone digs up the cable in the street. Basically, your mains pulls the power out of the batteries via the inverter and you don't get the lights don't go off. If you contrast that with a rotary set, basically a rotary set is a great big motor generator block. You under normal situations, the power, the AC power spins the motor generator up. And at the same time as it's spinning it up, you pull off the alternator side of it, an AC supply which goes to the load. But what can happen is you you have some sort of flywheel effect and you have some other various magical black boxes which each of the vendors have their own version of. And basically in the event of a failure, the inertia would carry it through for a certain amount of time. And then what happens is on the same shaft as the motor, the electric motor alternator combination, you have a generator. So you have a diesel engine and you have a clutch. So as soon as the main failure is sensed, you start the diesel engine up. And as soon as the diesel engine has got enough momentum to take over, the clutch shuts and you then start taking over. The clutch starts running the motor generator assembly rather than the AC motor was running it. These can get very, very complicated, very, very quick. And one of the things that really can be a big issue with these is the way that they're sold and the fact that they're, you know, these unlikely UPS, static UPS. Normally a drops or a dynamic rotary UPS has a short break and a no break rating. So basically this might be a, you know, 2.5 MVA UPS drop set. But what it will have is it will only have maybe a 2.2 or maybe even less than 2, maybe even 2 MVA no break capacity. And then the rest will have a short break. And the reason for that is basically what happens is it until the generator. So the diesel engine has taken over running the motor generator block. You're using the inertia to spin it down. And as the power is pulling the alternator and taking load off the alternate, it's slowing down, slowing down. And, you know, what you've got to do is if you get, if you leave it too long before the, if you've got too much load on it and you're pulling it off too quickly, you'll actually go to an under voltage situation quickly. So you've got to be very careful. The other thing is the rating that they normally sell these at in the catalog. They'll normally sell you these as a 0.8 power factor rating, which is just historical. That's what they used to use. 0.8 was everyone's favorite power factor. If you use them to supply a data center load, which has a unity power factor or a 0.95 power factor, then you need to consider that because, you know, they will rate this. The catalog rating is nearly always at 0.8 power factor. And I know situations where a combination of these factors have meant that these are, you know, people have ended up in court, shall we say. The only other thing I just talk about is the battery, typically five minutes. It might be up to 10 minutes. Now, obviously everyone wants to use lithium ion batteries. Lithium ion batteries bringing with them, obviously their own fire hazards. I think everyone is very, very aware of that. And they need very, very careful treatment in terms of what you're doing to actually accommodate the batteries and and how you are going to deal with any potential fire. You know, God forbid should happen. So I've been speaking probably a bit longer than I thought, but I still got a little bit to go. But, you know, keep on going, go to mechanical. Mechanical, right. OK, this is just a cut from what ASHRAE would say where you need to be with in that little recommended zone in the center. And then the A1, A2, A3, A4 are different levels of equipment. You know, A1 is enterprise grade equipment. And, you know, A4 is probably just PCs and lower important equipment. So what they're saying is humidity as long as it's non condensing is not really a drama. And obviously what's happened over the years is we've started to creep up from it being just a 1820 C environment, which was we thought we were keeping everything cool and sweet. And, you know, and also keeping it nice for people to work in temperatures are getting up and up and up. Obviously this is one of the key sources that, you know, we see cooling the power. Not much you can do about the power that these processes use, but obviously there's a lot you can do about the cooling. Interesting. If you look at Intel microchips spec sheets for, you know, processors, what they will say is that the actual performance of the actual GPU graphics processing unit doesn't start to throttle, meaning that they don't start degrade until the chip temperature reaches 80 degrees C, which I found very interested. And then I just got this thing recently where I cut from a website from NVIDIA and they're saying their GeoForce processor graphics cards, basically what type, which is the normal operating temperature. And they're saying that basically it can run up to 90 C and the maximum operating temperature is 105 C. And they're saying, you know, to operate reliably, reliably up to that limit. So you then start thinking, gosh, what are we doing? Why, you know, why are we cooling these as much as we are cooling them? Obviously, you know, we need to have people to go and access the equipment. But, you know, clearly I think we as an industry have been over cooling for some time now. How historically have we done it? Originally, you had the famous crack unit or cry unit, depending on what you want to call it. And they were just basically drawing the warm air back from the computers, circling it through a downflow unit and typically blowing it through a plenum raised floor, which was also used as a cabling route. And basically you pressurized the floor, put the outlets where you needed the load and then the hot air went back to the cabinet. A little while ago, some clever people realized that, you know, the equipment cabinets, they took air in at the front and they, you know, the cool air in at the front and they blew the hot air out the back. And they thought, well, oh, why don't we believe it or not, it seems so logical now that, you know, I can remember when people didn't do it, but why don't we actually spin the cabinets round? So we have all the inlets facing each other and all the outlets facing each other. So what you then have is a situation which is roughly on this on this diagram here where the inlets are all fed with cold air and all the outlets come to into common hot air discharge. So this was termed hot and cold aisles, the aisle being the walkway between the cabinets. So obviously when people started getting really sensible about cooling and not wanting to, what used to happen is some of the cold air that went into the cold aisle was actually went over the back of the cabinet, around the side of the cabinet and mixed into the hot air in the hot aisle. So the air that was actually brought back to the crack unit was actually cooler than it ideally should have been because, you know, you had some areas where the equipments were putting out a lot of, you know, doing a lot of work and this would have been diluted by spillage of cold air. So you weren't really getting an accurate balance between the load that you needed and what the cooling units were doing. So what people started doing is containing the one or the other aisles and the term hot aisle or cold aisle containment was derived. So this top figure there is basically aisle containment. I'm not sure whether it's hot aisle or cold aisle. So it's actually hot aisle when I look at it. So, and hot aisle is probably the most common. So what that means is the cold air is coming to all these grids. You see the three grill, sorry, go back. The three, the series of three floor grills there. That's the cold air coming into the front of the cabinets and it's on the other side. There's some cold air coming into the other cabinets. And the hot aisle is all going straight out into that chimney back into the ceiling area. And then it's been directed over the ceiling void back to the terminal cooling unit where it will get cooled and then sent back out again. So there you get a much better representation because you're not getting any unplanned dilution of the air. So basically you're not just chucking cold air out to do, to do, to do nothing. So more and more as time goes on, this is at that top picture is a sort of halfway house because you've got a floor void, a floor plenum and you've got a ceiling plenum. As time's gone on now, what people are doing is getting rid of the more and more, the crack units and replacing them by fan walls. So what's a fan wall? Basically, it's just a whole series of fans blowing through coils at one end of on one elevation, the elevation that's perpendicular to the aisle. And they blow cold air down the the cold aisles, basically. So the problem with this all becomes when you're getting to high cabinet loads and I'm talking, you know, anything more than about 20 kilowatts of cabinet, you're going to end up moving a huge volume of air. I mean, I once did a presentation to DCD in Singapore and I had a slide which was showing string vests and people were going like, what you got that slider for? I said, well, yeah, string vests work and keep you warm in winter in back in the UK because they've got an air pocket in them and they keep you cool in summer because they've got an air pocket in them. And why is that as well? Because air is a pretty good conductor and it's pretty lousy, pretty good insulator and a pretty lousy conductor. And yet for years and years and years we've been using this lousy conductor, good insulator to take heat away from equipment. And it just in my mind, it is starting to beg a belief. And so what we're we're moving on to now is liquid cooling, which I think is going to be more and more the way we see things going. So liquid cooling, you're going to have two slides with the same pictures and the same titles, but the slightly less words are slightly different. So some sorts of liquid cooling, top one is cold plates. Basically, this little white device there is designed to be like a heat sink. It sits on top of whatever the GPU, the CPU or the memory in the computer is and you run a liquid through it. It could be, you know, some form of purified water or it could be some form of a silicon oil or it could be a phase change liquid. There's different liquids being used, but the idea is that you liquid cool the chip where the heat is itself. And if you go back to a couple of slides ago where I said that the chip, you know, can run happily up to 105 C. If you pump a liquid through that most places in the world, you probably have to throw that onto the roof and ditch it into a dry with a dry cooler rather than using a cooling tower. So, and you won't need any chiller. You know, even in most of this part of the world, I don't know what the average summer design day temperature would be probably in most places, not much north of 40 C. And if you're running water through these that, you know, in the seventies or liquid run at seventies or eighties, you should still be able to ditch that through a dry cooler. Obviously, ditching the heat is not the best thing to do with it. It would be much better if we could use it for some sensible purpose. But at the moment, you know, I would say the vast majority of data centers out there are using either a cooling tower to reject the heat or they're using some form of a way of rejecting the heat. Occasionally you'll see someone using it to heat a swimming pool or, you know, heat some houses in Sweden, but it's still nowhere near where it should be, where we should be. The next sort of one is perhaps a little bit more conventional rear door heat exchanges. I think this is probably a very useful halfway house at the moment. And here what you have is a big coil on the back of the equipment cabinet. You reject the heat through the equipment cabinet in as you would do before, but this time before it gets blown into the hot aisle, because in this case you wouldn't have a hot aisle per se. You would have this big thin coil and you'd run some liquid through that and you would remove the heat from coming off the equipment in that liquid in that that that thin coil in the back. And so this may or may not have fans in it and you can use different fluids again. Quite often what you'll do is run a return pair using a cool water rather than a chilled water above the cabinets and drop off to the to these doors. And sometimes you use these to actually provide some in rule in row cooling as well. So you'll over cool them so that you get some temperature, some temperature modulation and in row just general cooling as well as just the heat from from these as well. The benefits to my mind of these as you can get you don't you're not moving massive quantities of air around as you were. And obviously then the fan loads are you end up with the typically the pump loads reduced because the delta P over these coils are much less than a delta P over a car or something like that. And sometimes you might put a plate and frame exchanger is an interposing frame exchanger between these and the main area so that if you drop the, you know, the liquid contents of this, you have a limited leak rather than dropping the entire contents of the building. Yeah, and I see these being used more and more you can because you don't with the fan wall or the cry units, you know, because you you might have extra wide cold aisles to get the volume of air that you need down them. So what you can even get is with the in some instances where the volume of air you need to push down the cold aisle is so high that if you push it down a typical two tile wide, but it used to always be two tile 1200 wide a cold aisle and the velocity past the first few cabinets on on the line is such that you get a venturi effect and you can stall the fans in the equipment on the first few cabinets. So you end up making the the, you know, the cold aisle wider as well as that you'll need some circulation space for round and you need maintenance space. So if you can get rid of all that and you can use the aisle between the two rows of equipment and you don't need any containment to actually use as a maintenance space for the rear door heat exchanger, you can actually either narrow down the size of your data hall or have more capacity in a given size of floor area. These have been used as a retrofit as well. I think they're as you can probably hear from my enthusiasm that more and more popular immersion cooling is basically where the equipment sort of turned on its back and full and filled up with a non conducting liquid. I haven't seen these being used. I think everyone's still a little bit finds them a little bit novel. There was one guy was talking to me about them and saying because they use the oil, you know, quite often use some form of silicon oil in there that if you had a fault with it, you'd end up frying your servers deep frying your servers. I think that was a little bit tongue in cheek, but anyway, that's that. So just on the cold plate. So they're not that popular. The reason why is the guy who's supplying the servers is not the guy who's done the MEP installation. And there's some concern that you have to get your tenant or your customer to buy servers that can can have these cold plates fitted to them. There are more and more, you know, companies that will will do that. Now, that company Cool IT, they will they have got Hewlett Packard and Dell servers certified using their their chips that they're cold plates. So and they've done large scale data centers with them. So, yeah, the obvious. The other thing is for years and years, people have been scared about putting any liquids, particularly water into data holes. So you do get a little bit of not on my watch, but you know, you get that with pretty much any of these solutions. The rear door heat exchanger said that they are gaining some popularity. Immersion, never seen them. So I think I've gone through the benefits of these liquid cooling systems before. I realize I'm overrunning a little bit, but I'll speed through just very quickly some personal foibles. Quite often uptime Institute tier three requires you to the systems concurrently maintain that. Sorry, I shouldn't say concurrently maintainable. It should concurrently maintainable. Sorry, not concurrently maintainable. So some people have interpreted this or they interpret that meaning that for every valve that you have, you should be able to take the valve out without, you know, without downtime. So what they're then saying is you'll need two valves. So this this growing trend, which I pull my hair out and fight like tooth and nail where everywhere you have a valve, you put two valves and I just can't say I can't understand why people believe that if you've got one valve in a circulation system and it's broken for some reason that if you had another spare valve next to it, that that won't break for some reason. And if you look at, you know, as I did recently, like a 750 mil diameter butterfly valve for some ring main on the roof of a building, the mean time between failures measured by some Norwegian oil experts is 21 years. So you think, well, you really need to put two of those in anyway, that's that the other thing just on hydronics is most of the data centers that we see use ring mains for the distribution. So you know, yeah, high, you know, over volume, low pressure drop ring circuits, and they might have a horizontal and a vertical. So you might depending on where your terminal equipment is, you have typically a horizontal circuit and then you'll have vertical circuits feeding the different floors. And the whole idea of this is you can take a section of the ring main out without to resolve a leak and you've got alternative routes available. And another benefit of this means that the low pressure drop means that, you know, whichever way you're running, because the majority of the resistance is always in the terminal unit, you know, you're not going to, it's going to tend to self-balance and you're unlikely to have problems with pump performance because, you know, really, you know, all the resistance is going to be in the terminal compared to the high pressure, the low pressure ring mains. Very quickly on fire, fire mostly use pre-action sprinklers as just a sprinkler system where you don't fill it with water, you keep it full of, you know, the actual range pipe is full of compressed air. And what happens is in the event of two knocks in the fire system being picked up, you'll evacuate the compressed air from the line and allow the line to charge with the sprinkler water. The sprinkler won't go off until the sprinkler bulb is burst by, you know, the heating of the bulb and, you know, so you don't get water into the data hole. You'll only get water into the data hole if obviously the, you get, you know, two knocks on the fire, you get rid of the compressed air in the system and you fill it with water and the sprinkler goes. Most of the data centers used to have data extinguishing systems, they're not as much having extinguishing systems now than normally just reserve them for the switch rooms. You'll also have different sorts of gases you can use. And again, you can either have a local box store per area or centralized bottle store quite often. We're finding that people have a centralized bottle store and they look at the worst case discharge and have allowed for two, two worst case discharges through the whole building against a double knock system. And obviously you talked about generators before the lots of generators on these jobs. You're not really going to do any good with them with a gas. You don't really want to put through the sprinkler system in there. So quite often the generator enclosures or generator rooms are fitted with a fine mist sprinkler system and they would use deionized water in under pressure in some sort of pressure store, you know, pressure vessel. And then just put in is a fine mist in the event of a fire. Lots of other things you could talk about about the fire services, but that's generally the main thing. I just wanted to talk about cabling types. Obviously the data center's got huge amounts of cabling. It used to be unshielded twisted pair, CAT 6. Now I like to be CAT 7, shielded twisted pair, some performance figures. Typically what you're saying is the copper cable is limited to 80 meters, typically maybe a little bit more. You might get 100 meters out of it depending on what you want, but most people say 80 is a safe thing. Fiber optic cable more and more used now. Obviously always in pairs as a transmit received pair. Multi mode used within the building. And then single mode used for campuses and, you know, longer distances. Just as a sort of, you know, talk about some of the architecture. The hierarchical system, which is what buildings used to be designed on, which was where you had a, you know, multi tinted building where you started off with an access layer. You went up to an aggregation layer, then you went to core layer and you had, you know, basically the flow of information was really upwards or downwards. So it was basically information being sought from the access layer and then taken back out the core layer or vice versa. But more and more, you're now getting a spine and leaf arrangement in data centers where you actually consolidated the layers down a bit. And the whole point of the spine and leaf is that you get information transmitted across the data center, you know, east to west rather than north to south, as they call it much, much quicker. Again, you could run a separate whole separate seminar on these. And then just briefly onto security, realizing that I'm over time. It's very clearly a huge issue. Montreal Authority of Singapore requires any data center where there's a financial transaction to take place to do threats and vulnerabilities risk assessment, which goes through. You're right from, you know, can someone drive a car in with a bomb in it of a certain number of kg of TNT right through to have you checked where you're getting your staff from, et cetera, et cetera. So typically with these data centers, they have a low access. They have it in tiered zones. They have parameters. So you have a low access area. Even their low access area is probably similar to you. You'd have an airport air site. Then you have a medium layer where you basically have more authority. And then you have the high level where you're into really, you know, card in, card out, maybe even, you know, biometric controls, iris controls, something like that. And, you know, everything on CCTV and very intelligent CCTV where you can do all sorts of tricks with setting up lines on the seating, you know, pixel lines that people can't cross. People can't loiter. Any movement will, you know, zoom the camera into it. You know, cameras to doors, et cetera, et cetera. So more and more and more of ELV systems. And then obviously outside the building, you don't really want people getting in as the first line of defense. So there's nearly always crash barriers capable of stopping a car laser intrusion detection. So they can basically scan the outside of the perimeter and they'll pay, you know, get a breach of the security fence within a meter. You know, CCTV cameras are giving you, you know, very high density images, you know, both day and night, vehicle nameplate record, number plate recognition, all the, you know, things that goes on and on. I would say that they're short of prisons. They're probably the most secure buildings that you'll ever come across. I think this is pretty much the commission, the finishing commissioning. Obviously, as engineers, you know, we realize that commissioning is really where the rubber hits the road for one of a better series. All of the data centers I've worked on have the independent commissioning authority. And this person is really only there to make sure that the building is commissioned properly. They're not the designer. They're not the contractor. They're independent. Typically, they'll use this five stage process where any equipment that's made off site will be factory acceptance tested. When it comes to site, it will be unpacked and checked on site to make sure it hasn't been damaged. Once it's actually been inspected and put into service, it will be checked as its single unit. Then it will be checked as a part of a subsystem. And then level five will be an integrated system test where all of the systems are tested together. And quite often a number of the customers that we work with will use a physical label on every bit of equipment. So they may follow this color scheme and you may actually have go to a piece of equipment and say, right, it's got it's got no level three green tag on it. We can't power it up. And then it is being powered up a level. We had the level three test green tag on it. We've powered it up. We'll now test it locally. And then, you know, level five will go when we test it is the whole thing together. So very, very important. Really, that's just the conclusion to what I wanted to talk about today. I hope it's been of some interest. I recognize I've gone on probably a little bit longer than I intended to. As I said before, I'm really just from a point of view, a person, you know, working in data centers and seeing data centers and seeing what commonly happens. It's not at all prescriptive or recommended or anything like that. It's just really, you know, I guess what I've seen, I guess. Okay. So that's it from me. [01:06:22] Speaker 3: I don't want to know whether I want to hand it back to Matt. Are you there, Matt? Yes, question time. [01:06:40] Speaker 2: All right. [01:06:41] Speaker 1: So we've got a question from Adnan. Would the 1.4 PUE be dependent on the environment the data center is located? And what would be an acceptable PUE in Australia for an example? [01:06:54] Speaker 4: Good question. No, it doesn't seem to be. The only thing I would say is, for example, I would say probably 1.4 was probably getting a bit high in Australia because Australia is considered developed and it's, you know, fairly well serviced generally. You know, where we would say a 1.4 is maybe somewhere like India where water may be a scarce resource and that that would typically force you to go to a, you know, a air cooled chiller system rather than a cooling tower system. On the other hand, if you go to somewhere like Norway or Sweden or, you know, the Nordics generally where, you know, you should be able to cool it with fresh air most of the year round, you 1.4 would be pretty low, you know, pretty, pretty poor. If you'd hope to get down to 1.1, you'd hope to get down to 1.1, 1.2 quite often. And I think one of the questions that, you know, people ask is, could you ever get below one or could you get one as a I don't think you could ever get below one or but someone I think I think actually someone told me that went for an interview with one of the big cloud providers and they said, could you ever get a below one? And I think logically would say that no, you couldn't because, you know, how could you have the power feeding the building being less than the power feeding the data center, the data equipment. But I'm not sure that that's the case, you see, because what I was wondering was whether you couldn't actually recover the heat and then generate power from the heat. And this brings me back to a paper that I read many, many years ago when I was in college, and it was talking about solar power. And this was saying that there was a, and this is going back many years, there was a solar power scheme put in place in Saudi Arabia. And what they did in the solar power scheme is they used the solar heated water to actually boil a refrigerant. And then they used the refrigerant, expanded the refrigerant gas, the boiled refrigerant gas through a turbine, and they connected the turbine shaft to a pump. And they used that to pump their water through their desalination plant. So my view was that you could just, instead of using the solar power to heat the water, you could use the waste heat from the computer room. And then instead of actually coupling the refrigerant turbine drive to a pump, you could put it onto an alternator. And therefore you could actually potentially generate power. And therefore you could actually use some of the waste heat to generate power. And therefore the, you know, you could get a PUE less than one. Never seen it done. Never seen anyone advertising it, but it was just a thought I had. [01:09:59] Speaker 1: I guess it's a sort of a utopia of energy performance that we can aim to target in the future. [01:10:07] Speaker 4: Yeah, but I couldn't say any, maybe there's someone on the call that can, but I couldn't say any logical flaws in the logic. If there was indeed this desal plant in Saudi that was using solar heated water to drive a refrigerant turbine, why not? [01:10:28] Speaker 1: Yeah. So I guess in line with that, I've got a couple of questions. So what measures are being considered or implemented to data centers to achieve net zero? [01:10:40] Speaker 4: Well, I would say that most of the data center companies that we work with have got a huge environmental department that look at pretty much everything that they do. So, you know, I think they're well aware of the fact that they're consuming a lot of power. And therefore, you know, nearly all of them will will buy power from green sources. They will try and locate near wind farms or hydro. Or so if you look in, you know, somewhere like Sweden or Norway, if you can use all your hydro to all your powers, hydro power, green power, and then you're doing very little cooling. And then you're heating the university block next door or something like that, then then you're somewhat near there. So I've seen all sorts of other things, obviously analysis of everything that goes into the data center. But there was even recently I was at a seminar where there was a talk of a data center operator who had thought they'd got down to net zero or very close to net zero by some of the measures that I just talked about, and then realized that they ran their generators once a month to test them. And that was blowing a hole in it. So what they then were doing was using a bigger pack type of scenario where they got the generator, got rid of the generators and they put a massive battery pack in place and that their power reserve wasn't a generator. And it wasn't, you know, an oil fire generator anywhere. It was a, you know, a battery pack. The other thing I think will eventually, again, if you look at somewhere like Iceland where, you know, power, geothermal power is relatively inexpensive. I think, you know, if you could use hydrogen and maybe then start to use hydrogen as a fuel for some of these systems, that might be another way forward. But, you know, I think that all the big players have got, you know, far better brains than mine looking at looking at this, you know. [01:12:49] Speaker 1: So I guess, I mean, I'll derail this a bit, but I mean, you know, you look at tools like Greenstar in Australia where they're, you know, pushing the industry towards electrification and getting rid of diesel generator systems. Do you see that feasible long term for these type of facilities at a large scale? [01:13:09] Speaker 4: Yeah, you know, I think what needs to happen is there needs to be a seed change in battery technology. And I think we will see with the likes of people like Elon Musk and Tesla massive developments in battery technology. I think, you know, I think we're already seeing massive improvements in solar power. And if you look at, you know, I think Elon Musk has said that he could power the whole of the United States with solar power in an area somewhat smaller than Utah. You know, and I think he's actually sketched on maps the amount of space you would need to solar power the whole world. So I think, yes, you know, it's probably a little bit academic, but I think it will happen. I think I don't think I'll see it, but I think, you know, my kids might. [01:14:03] Speaker ?: Cool. [01:14:04] Speaker 1: Now question from Chanel. How does the Melbourne climate benefit immersion technology? [01:14:12] Speaker 4: I don't know is the question. I'm not familiar with it. The only thing I know about the Melbourne climate is it was made famous in that crowded house song. Four seasons in one day, but no, I don't know. But all I would say is that if you look at liquid cooling and you look at back, I think fundamentally what I was saying about. I think the great untold story is that processes can take a lot higher temperatures than we've given them credit for. So I think that basically what you could do is probably take liquid from a immersion system and put it through some form of direct cooler dry cooler system and, you know, rather than have say cooling towers and refrigeration. That would be my guess. [01:15:00] Speaker ?: That would be my guess. [01:15:00] Speaker 4: But I mean, I'd be happy to hear if it, you know, if he's got some first-hand experience of it that he's happy to share. [01:15:08] Speaker 1: Sure. I guess. So I get on from that is another regional question. So what's the delta T we would want to use for choosing crack units in Perth? And if you're not familiar with Perth, it's quite hot, dry climate. [01:15:23] Speaker 4: Well, I think what I would say is I probably wouldn't use crack units because I think what I would do is I'd use a rear door heat exchanger scheme, you know, and then I would look at what the ambient temperature was and whether I could actually run. Depending, you know, what some operators have done is they have done a situation where they pre-cool, they use the cooling, they use cooling towers and they use chillers. But what they do is they use the cooling tower to pre-cool the water that they're feeding to the chillers. So they use high temperature chillers, not low temperature chillers. So, you know, my guess is that, you know, 1820 is a flow and maybe, you know, seven or eight delta T on that. And then what they do is they use the hot water, say the in-row coolers, the rear door plate exchangers, and then they would use that, put that through a plate exchanger. On the cooling tower system, and then the pre-cooled water that they, you know, the return water they got that pre-cooled off the plate exchanger on the cooling towers, they then put into a chiller, into a high temperature chiller. But yeah, off the top of my head, as I said, I'm not a design engineer. I would need to sit down and think about what I would choose for that part of the world as a flow and return temperature. But I personally wouldn't use cracks. I think the issue with them is that you're going to need space. You're going to use N plus 25% units, so you're going to have a lot of motors on there. You're going to have a, normally you have a situation where you'd put them both sides of a hole, so that if you have a failure on one side, you can get it from the other side. Then you have a situation where you will normally have some sort of plenum to equalize the pressure so that if you have a failure along the length of the row of cracks, you can compensate with it. And then you nearly always need to have access and more and more nowadays for security and because they don't want a guy with a pair of wrenches into a white space in case he's a Russian spy and he's got a thumb drive that he's going to try and plug into something. What they'll try and do is they'll try and keep the people out so they'll have a maintenance corridor behind the cracks. So it's not unusual. I've seen quite often you have dead space behind the crack is a maintenance corridor and equalizing pressure plenum in front of the cracks and then you blow into the room and then you have a space that allows you to get round the ends of the racks to circulate round the end of the racks. So you've really then got three loads of circulation space and the crack space, whereas if you look at just putting in something like a rear door heat exchanger scheme, you get rid of all that. So you can have either a much smaller building and you're not blowing a lot of air around because if you've got cracks, you've got to get the air back to the cracks. So you're going to have to have either a big plenum or wide cold aisles and all going to go on your end cost of your building. Having said that, the end cost of your building is probably only going to be 20% of the overall cost of the data center. So it's not going to be like, you know, in a hotel or, you know, office block where saving, you know, CSA cost is going to be a huge saving to you. It's not going to be the same saving as it would be elsewhere, but it still will be saving. And it obviously is the green issue of all the materials as well. Fantastic. [01:19:08] Speaker 1: And last question, does ASHRAE guidelines take precedence over local standards regulating buildings in general in the case of data centers? [01:19:19] Speaker 4: I would say not. I would say, you know, at the end of the day, you're always going to have to face a guy down at County Hall, aren't you? You know what I mean? Nearly always everywhere you go, you're going to have to get a chop on it and you're going to have to get someone to sign it off to approve use on it. So, no, the only thing I've found is that most times if you talk to people in the local authorities and you explain to them what you're doing and where you've got the information from, they'll be okay with it, you know. But, you know, I would never ever condone short circling local regulations in any shape or form because they're the guys who are going to end up, you know, certifying it, you know. And there may be some foible of that area that, you know, ASHRAE may not be in its infinite wisdom aware of, you know. [01:20:03] Speaker 1: Awesome. Well, look, that's the end of the question, so I think we'll wrap it up there. But just want to, on behalf of everyone, I just want to really thank you for your time, Chris. It was a fantastic presentation, very, very informative. Actually, I've got a, someone's raising their hand now, so if you've got a question, please write it down. But otherwise, look, it was really, really awesome, Chris. I really appreciate it. [01:20:30] Speaker 4: Well, as I say, you know, it's difficult to hit the right level. I hope I've not bored anyone to tears or, you know, in any way, you know, missed something that someone would have been like to learn. But anyway, you probably know that I love the business and it's a pleasure, if you can, to, you know, to share some of your experience with other people. It's something I think that as engineers, we should be doing as a matter of course, you know. Awesome. [01:20:54] Speaker 1: Well, I guess for everyone online, this has been recorded. So this will get distributed via the SIPC website, I think within a month. But if you have any questions, send them through to SIPC and we'll pass them on to you, Chris, and maybe we can. Maybe we can intro people online and they can contact you directly if they want to discuss further. Yeah. [01:21:16] Speaker 3: Okay. [01:21:17] Speaker 1: Thanks very much. [01:21:18] Speaker 3: Thanks very much, Chris. Thanks very much, everyone. Have a good night.