As 2020 ushers in a new decade, the forces shaping data center decisions are extending compute resources to new places.
With the challenging goals of speed, agility, and efficiency, enterprises and service providers alike will be seeking new balance between the need for low latency and optimal utilization of workload placement. Hybrid models will therefore include more distributed, confined, and modular data centers at or near the edge.
These are but some of a few top-line predictions on the future state of the modern data center design. The next BriefingsDirect data center strategies discussion with two leading IT and critical infrastructure executives examines how these new data center variations nonetheless must also interoperate seamlessly from core to cloud to edge.
Here to help us learn more about the new state of extensible data centers is Peter Panfil, Vice President of Global Power at VertivTM, and Steve Madara, Vice President of Global Thermal at Vertiv. The discussion is moderated by Dana Gardner, Principal Analyst at Interarbor Solutions.
Here are some excerpts:
Gardner: The world is rapidly changing in 2020. Organizations are moving past the debate around hybrid deployments, from on-premises to public clouds. Why do we need to also think about IT architectures and hybrid computing differently?
Panfil: We noticed a trend at Vertiv in our customer base. That trend is toward a new generation of data centers. We have been living with distributed IT, client-server data centers moving to cloud, either a public cloud or a private cloud.
But what we are seeing is the evolution of an edge-to-core, near-real-time data center generation. And it’s being driven by devices everywhere, the “connected-all-the-time” model that all of us seem to be going to.
And so, when you are in a near-real-time world, you have to have infrastructure that supports your near-real-time applications. And that is what the technology folks are facing. I refer to it as a pack of dogs chasing them — the amount of data that’s being generated, the applications running remotely, and the demand for availability, low latency, and driving cost down as much as you possibly can. This is what’s changing how they approach their critical infrastructure space.
Gardner: And so, a new equilibrium is emerging. How is this different from the past?
Madara: If we go back 20 years, everything was centralized at enterprise data centers. Then we decided to move to decentralized, and then back to centralized. We saw a move to colocation as people decided that’s where they could get lower cost to run their apps. And then things went to the cloud, as Peter said earlier.
And now, we have a huge number of devices connected locally. Cisco says by late 2020 that it’s going to have 23 billion connected devices, and over half of those are going to be machine-to-machine communications, which, as Peter mentioned earlier, the latency is going to be very, very critical.
An interesting read is Michael Lewis’s book Flash Boys about the arbitrage that’s taking place with the low latency that you have in stock market trading. I think we are going to see more of that moving to the edge. The edge is more like a smart rack or smart row deployment in an existing facility. It’s going to be multi-tenant, because it’s going to be able to be throughout large cities. There could be 20 or 30 of these edge data center sites hosting different applications for customers.
This move to the edge is also going to provide IT resources in a lot of underserved markets that don’t yet have pervasive compute, especially in emerging countries
Gardner: Why is speed so important? We have been talking about this now for years, but it seems like the need for speed to market and speed to value continues to ramp up. What’s driving that?
Moving to the edge, with momentum
Panfil: There is more than one kind of speed. There is speed of response of the application, that’s something that all of us demand — speed of response of the applications. I have to have low latency in the transactions I am performing with my data or with my applications. So there is the speed of the actual data being transmitted.
There is also speed of deployment. When Steve talked earlier about centralized cloud deployments in these core data centers, your data might be going over a significant distance, hopping along the way. Well, if you can’t live with that latency that gets inserted, then you have to take the IT application and put it closer to the source and consumer of the data. So there is a speed of deployment, from core to edge, that happens.
And the third type of speed is you have to have low-first-cost, high-asset-utilization, and rapid-scalability. So that’s a speed of infrastructure adaptation to what the demands for the IT applications are.
So when we mean speed, I often say it’s speed, speed, and speed. First it’s the data speed, then deploying fast, and then at scale at business-friendly cost and reliability.
So when we mean speed, I often say it’s speed, speed, and speed. First, it’s the data IT. Once I have data IT speed, how did I achieve that? l did it by deploying fast, in the scale needed for the applications, and lastly at a cost and reliability that makes it tolerable for the businesses.
Gardner: So I guess it’s speed-cubed, right?
Panfil: At least, speed-cubed. Steve, if we had a nickel for every time one of our customers said “speed,” we wouldn’t have to work anymore. They are consumed with the different speeds that they have to deal with — and it’s really the demands of their customers.
Gardner: Vertiv for years has been looking at the data center of the future and making some predictions around what to expect. You have been rather prescient. To continue, you have now identified several areas for 2020, too. Let’s go through those trends.
Steve, Vertiv predicts that “hybrid architectures will go mainstream.” Why did you identify that, and what do you mean?
The future goes hybrid
Madara: If we look at the history of going from centralized to decentralized, and going to colocation and cloud applications, it shows the ongoing evolution of Internet of Things (IoT) sensors, 5G networks, smart cities, autonomous cars, and how more and more of that data is generated and will need to be processed locally. A lot of that is from machine-to-machine applications.
So when we now talk about hybrid, we have to get very, very close to the source, as far as the processing is concerned. That’s going to be a large-scale evolution that’s going to drive the need for hybrid applications. There is going to be processing at the edge as well as centralized applications — whether it’s in a cloud or hosted in colocation-based applications.
Panfil: Steve, you and I both came up through the ranks. I remember when the data closet down the hall was basically a communications matrix. Its intent was to get communications from wherever we were to wherever our core data center was.
Well, the cloud is not going away. Number two, enterprise IT is not going away. What the enterprise is saying is, “Okay, I am going to take my secret sauce and I am going to put it in an edge data center. I am going to put the compute power as close to my consumer of that data and that application as I possibly can. And then I am going to figure out where the rest of it’s going to go.”
If I can live with the latency I get out of a core data center, I am going to stay in the cloud. If I can’t, I might even break up my enterprise data center into small or micro data centers that give me even better responses.
“If I can live with the latency I get out of a core data center, I am going to stay in the cloud. If I can’t, I might even break up my enterprise data center into small or micro data centers that give me even better responses.”
Dana, it’s interesting, there was a recent wholesale market summary published that said the difference between the smaller and the larger wholesale deals widened. So what that says is the large wholesale deals are getting bigger, the small wholesale deals are getting smaller, and that the enterprise-based demand, in deployments under 600 kilowatts, is focused on low-latency and multi-cloud access.
That tells us that our customers, the users of that critical space, are trying to place their IT appliances as close as they can to their customers, eliminating the latency, responding with speed, and then figuring out how to mesh that edge deployment with their core strategy.
Gardner: Our second trend gets back to the speed-cubed notion. I have heard people describe this as a new arms race, because while it might be difficult to differentiate yourself when everyone is using the same public cloud services, you can really differentiate yourself on how well you can conduct yourself at speed.
What kinds of capabilities across your technologies will make differentiation around speed work to an advantage as a company?
The need for speed
Panfil: Well, I was with an analyst recently, and I said the new reality is not that the big will eat the small — it’s that the fast will eat the slow. And any advantage that you can get in speed of applications, speed of deployment, deploying those IT assets — or morphing the data center infrastructure or critical space infrastructure – helps improve capital efficiency. What many customers tell us is that they have to shorten the period of time between deciding to spend money on IT assets and the time that those asset start creating revenue.
They want help being creative in lowering their first-cost, in increasing asset utilization, and in maintaining reliability. If, holy cow, my application goes down, I am out of business. And then they want to figure out how to manage things like supply chains and forecasting, which is difficult to do in this market, and to help them be as responsive as they can to their customers.
Madara: Forecasting and understanding the new applications — whether it’s artificial intelligence (AI) or 5G — the CIOs need to decide where they need to put those applications whether they should be in the cloud or at the edge. Technology is changing so fast that nobody can predict far out into the future as far as to where I will need that capacity and what type of capacity I will need.
So, it comes down to being able to put that capacity in the place where I need it, right when I need it, and not too far in advance. Again, I don’t want to spend the capital, because I may put it in the wrong place. So it’s got to be about tying the demand with the supply, and that’s what’s key as far as the infrastructure.
And the other element I see is technology is changing fast, even on the infrastructure side. For our equipment, we are constantly making improvements every day, making it more efficient, lower cost, and with more capability. And if you put capacity in today that you don’t need for a year or two down the road, you are not taking advantage of the latest, greatest technology. So really it’s coupling the demand to the actual supply of the infrastructure — and that’s what’s key.
Another consideration is that many of these large companies, especially in the colocation market, have their financial structure as a real estate investment trust (REIT). As a result, they need to tie revenue with expenses tighter and tighter, along with capital spending.
Panfil: That’s a good point, Steve. We redesigned our entire large power portfolio at Vertiv specifically to be able to address this demand.
In previous generations, for example, the uninterruptible power supply (UPS) was built as a complete UPS. The new generation is built as a power converter, plus an I/O section, plus an interface section that can be rapidly configured to the customer, or, in some cases, put into a vendor-managed inventory program. This approach allows us to respond to the market and customers quicker.
We were forced to change our business model in such a way that we can respond in real time to these kinds of capacity-demand changes.
Madara: And to add to that, we have to put together more and more modules and solutions where we are bundling the equipment to deliver it faster, so that you don’t have to do testing on site or assembly on site. Again, we are putting together solutions that help the end-user address the speed of the construction of the infrastructure.
I also think that this ties into the relationship that the person who owns the infrastructure has with their supplier base. Those relationships have to build in, as Peter mentioned earlier, the ability to do stocking of inventory, of having parts available on-site to go fast.
Gardner: In summary so far, we have this need for speed across multiple dimensions. We are looking at more hybrid architectures, up and down the scale — from edge to core, on-premises to the cloud. And we are also looking at crunching more data and making real-time analytics part of that speed advantage. That means being able to have intelligence brought to bear on our business decisions and making that as fast as possible.
So what’s going on now with the analytics efficiency trend? Even if average rack density remains static due to a lack of space, how will such IT developments as high performance computing (HPC) help make this analysis equation work to the business outcome’s advantage?
High-performance, high-density pods
Madara: The development of AI applications, machine learning (ML), and what could be called deep learning are evolving. Many applications are requiring these HPC systems. We see this in the areas of defense, gaming, the banking industry, and people doing advanced analytics and tying it to a lot of the sensor data we talked about for manufacturing.
It’s not yet widespread, it’s not across the whole enterprise or the entire data center, and these are often unique applications. What I hear in large data centers, especially from the banks, is that they will need to put these AI applications up on 30-, 40-, 50- or 60-kW racks — but they only have three or four of these racks in the whole data center.
The end-user will need to decide how to tune or adjust facilities to accommodate these small but growing pods of high-density compute. They are going to need to decide how they are going to facilitize for that type of equipment.
The end-user will need to decide how to tune or adjust facilities to accommodate these small but growing pods of high-density compute. And if they are in their own facility, if it’s an enterprise that has its own data center, they will need to decide how they are going to facilitize for that type of equipment.
A lot of the colocation hosting facilities have customers saying, “Hey, I am going to be bringing in the future a couple of racks that are very high density. A lot of these multi-tenant data centers are saying, ‘Oh, how do I provision for these, because my data center was laid out for this average of maybe 8 kW per rack? How do I manage that, especially for data centers that didn’t previously have chilled water to provide liquid to the rack?’”
We are now seeing a need to provide chilled water cooling that would go to a rear door heat exchanger on the back of the rack. It could be chilled water that would go to a rack for chip cooling applications. And again, it’s not the whole data center; it’s a small segment of the data center. But it raises questions of how I do that without overkill on the infrastructure needed.
Gardner: Steve, do you expect those small pods of HPC in the data center to make their way out to the edge when people do more data crunching for the low-latency requirements, where you can’t move the data to a data center? Do you expect to have this trend grow more distributed?
Madara: Yes, I expect this will be for more than the enterprise data center and cloud data centers. I think you are going to see analytics applications developed that are going to be out at the edge because of the requirements for latency.
When you think about the autonomous car; none of us know what’s going to be required there for that high-performance processing, but I would expect there is going to be a need for that down at the edge.
Gardner: Peter, looking at the power side of things when we look at the batteries that help UPS and systems remain mission-critical regardless of external factors, what’s going on with battery technology? How will we be using batteries differently in the modern data center?
Panfil: That’s a great question. Battery technology has been evolving at an incredibly fast rate. It’s being driven by the electric vehicles. That growth is bringing to the market batteries that have a size and weight advantage. You can’t put a big, heavy pack of batteries in a car and hope to have it perform well.
It also gives a long-life expectation. So data centers used to have to decide between long-life, high-maintenance, wet cells and the shorter-life, high-maintenance, valve-regulated lead-acid (VRLA) batteries. In step with the lithium-ion batteries (LIBs) and thin plate pure lead (TPPL) batteries, what’s happened is the total cost of ownership (TCO) has started to become very advantageous for these batteries.
Our sales leadership lead sent me the most recent TCO between either TPPL or LIBs versus traditional VRLA batteries, and the TCO is a winner for the LIBs and the TPPL batteries. In some cases, over a 10-year period, the TCO is a factor of two lower for LIB and TPPL.
Where in the cloud generation of data centers was all about lowest first cost, in this edge-to-core mentality of data centers, it’s about TCO. There are other levers that they can start to play with, too.
So, for example, they have life cycle and operating temperature variables. That used to be a real limitation. Nobody in the data center wanted their systems to go on batteries. They tried everything they could to not have their systems go on the battery because of the potential for shortening the life of their batteries or causing an outage.
Today we are developing IT systems infrastructure that takes advantage of not only LIBs, but also pure lead batteries that can increase the number of [discharge/recharge] cycles. Once you increase the number of cycles, you can think about deploying smart power configurations. That means using batteries not only in the critical infrastructure for a very short period of time when the power grid utility fails, but to use that in critical infrastructure to help offset cost.
If I can reduce utility use at peak demand periods, for example, or I can reduce stress on the grid at specified times, then batteries are not only a reliability play – they are also a revenue-offset play. And so, we’re seeing more folks talking to us about how they can apply these new energy storage technologies to change the way they think about using their critical space.
Also, folks used to think that the longer the battery time, the better off they were because it gave more time to react to issues. Now, folks know what they are doing, they are going with runtimes that are tuned to their operations team’s capabilities. So, if my operations team can do a hot swap over an IT application — either to a backup critical space application or to a redundant data center — then all of a sudden, I don’t need 5 to 12 minutes of runtime, I just need the bridge time. I might only need 60 to 120 seconds.
Now, if I can have these battery times tuned to the operations’ capabilities — and I can use the batteries more often or in higher temperature applications — then I can really start to impact my TCO and make it very, very cost-effective.
Gardner: It’s interesting; there is almost a power analog to hybrid computing. We can either go to the cloud or the grid, or we can go to on-premises or the battery. Then we can start to mix and match intelligently. That’s really exciting. How does lessening dependence on the grid impact issues such as sustainability and conserving energy?
Sustainability surges forward
Panfil: We are having such conversations with our key accounts virtually every day. What they are saying is, “I am eventually not going to make smoke and steam. I want to limit the number of times my system goes on a generator. So, I might put in more batteries, more LIBs or TPPL batteries, in certain applications because if my TCO is half the amount of the old way, I could potentially put in twice as much, and have the same cost basis and get that economic benefit.”
And so from a sustainability perspective, they are saying, “Okay, I might need at some point in the useful life of that critical space to not draw what I think I need to draw from my utility. I can limit the amount of power I draw from that utility.”
I love all of you out there in data center design, but most of them are designed for peak useage. These changes allow them to design more for the norm of the requirements. That means they can put in less infrastructure, less battery, to right-size their generators; same thing on cooling.
This is not a criticism, I love all of you out there in data center design, but most of them are designed for peak usage. So what these changes allow them to do is to design more for the norm of the requirements. That means they can put in less infrastructure, the potential to put in less battery. They have the potential to right-size their generators; same thing on the cooling side, to right-size the cooling to what they need and not for the extremes of what that data center is going to see.
From a sustainability perspective, we used to talk about the glass as half-full or half-empty. Now, we say there is too much of a glass. Let’s right-size the glass itself, and then all of the other things you have to do in support of that infrastructure are reduced.
Madara: As we look at the edge applications, many will not have backup generators. We will have alternate energy sources, and we will probably be taking more hits to the batteries. Is the LIB the better solution for that?
Panfil: Yes, Steve, it sure is. We will see customers with an expectation of sustainability, a path to an energy source that is not fossil fuel-based. That could be a renewable energy source. We might not be able to deploy that today, but they can now deploy what I call foundational technologies that allow them to take advantage of it. If I can have a LIB, for example, that stores excess energy and allows me to absorb energy when I’m creating more than I need — then I can consume that energy on the other side. It’s better for everybody.
Gardner: We are entering an era where we have the agility to optimize utilization and reduce our total costs. The thing is that it varies from region to region. There are some areas where compliance is a top requirement. There are others where energy issues are a top requirement because of cost.
What’s going on in terms of global cross-pollination? Are we seeing different markets react to their power and thermal needs in different ways? How can we learn from that?
Global differences, normalized
Madara: If you look at the size of data centers around the world, the data centers in the U.S. are generally much larger than in Europe. And what’s in Europe is much larger than what we have in other developed countries. So, there are a couple of things, as you mentioned, energy availability, cost of energy, the size of the market and the users that it serves. We may be looking at more edge data centers in very underserved markets that have been in underdeveloped countries.
So, you are going to see the size of the data center and the technology used potentially different to better fit needs of the specific markets and applications. Across the globe, certain regions will have different requirements with regard to security and sustainability.
Even though we have these potential differences, we can meet the end-user needs to right-size the IT resources in that region. We are all more common than we are different in many respects. We all have needs for security, we all have needs for efficiency, it may just be to different degrees.
Panfil: There are different regional agency requirements, different governmental regulations that companies have to comply with. And so what we find, Dana, is that what our customers are trying to do is normalize their designs. I won’t say they are standardizing their design because standardization says I am going to deploy exactly the same way everywhere in the world. I am a fan of Kit Kats and Kit Kats are not the same globally, they vary by region, the same is true for data centers.
So, when you look at how the customers are trying to deal with the regional and agency differences that they have to live with, what they find themselves doing is trying to normalize their designs as much as they possibly can globally, realizing that they might not to be able to use exactly the same power configuration or exactly the same thermal configuration. But we also see pockets where different technologies are moving to the forefront. For example, China has data centers that are running at high voltage DC, 240 volts DC, we have always had 48-volt DC IT applications in the Americas and in Europe. Customers are looking at three things — speed, speed, and speed.
And so when we look at the application, for example, of DC, there used to be a debate, is it AC or DC? Well, it’s not an “or” it’s an “and.” Most of the customers we talk to, for example, in Asia are deploying high-voltage DC and have some form of hybrid AC plus DC deployment. They are doing it so that they can speed their applications deployments.
In the Americas, the Open Compute Project (OCP) deploys either 12 or 48 volts to the rack. I look at it very simply. We have been seeing a move from 2N architecture to N plus 1 architecture in the power world for a decade, this is nothing more than adopting the N plus 1 architecture at the rack level versus the 2N architecture at the rack level.
And so what we see is when folks are trying to, number one, increase the speed; number two, increase their utilization; number three, lower their total cost, they are going to deploy infrastructures that are most advantageous for either the IT appliances that they are deploying or for the IT applications that they are running, and it’s not the same for everybody, right Steve?
You and I have been around the planet way too many times, you are a million miler, so am I. It’s amazing how a city might be completely different in a different time zone, but once you walk into that data center, you see how very consistent they have gotten, even though they have done it completely independently from anybody else.
Consistency lowers costs and risks
Gardner: A lot of what we have talked about boils down to a need to preserve speed-to-value while managing total cost of utilization. What is there about these multiple trends that people can consider when it comes to getting the right balance, the right equilibrium, between TCO and that all important speed-to-value?
Madara: Everybody strives to drive cost down. The more you can drive the cost down of the infrastructure, the more you can do to develop more edge applications.
I think we are seeing a very large rate of change of driving cost down. Yet we still have a lot of stranded capacity out there in the marketplace. And people are making decisions to take that down without impacting risk, but I think they can do it faster.
Standardization helps drive speed, whether it’s normalization or similarity. What allows people to move fast is to repeat what they are doing instead of snowflake data centers, where every one is different.
Peter mentioned standardization. Standardization helps drive speed, whether it’s normalization or similarity. What allows people to move fast is to repeat what they are doing instead of snowflake data centers, where every new one is different.
Repeating allows you to build a supply base ecosystem where everybody has the same goal, knows what to do, and can be partners in driving out cost and in driving speed. Those are some of the key elements as we go forward.
Gardner: Peter when we look to that standardization, you also allow for more seamless communication from core to cloud to edge. Why is that important, and how can we better add intelligence and seamless communication among and between all these different distributed data centers?
Panfil: When we normalize designs globally, we take a look at the regional differences, sort out what the regional differences have to be, and then put a proof of concept deployment. And out of that comes a consistent method of procedure.
When we talk about managing the data center effectively and efficiently, first of all, you have to know what you have. And second, you have to know what it’s doing. And so, we are seeing more folks normalizing their designs and getting consistency. They can then start looking at how much of their available capacity from a design perspective they are actually using both on a normal basis and on a peak basis and then they can determine how much of that they are willing to use.
We have some customers who are very risk-averse. They stay in the 2N world, which is a 50 percent maximum utilization. We applaud them for it because they are not going to miss a transaction.
There are others who will say, “I can live with the availability that an N+1 architecture gives me. I know I am going to have to be prepared for more failures. I am going to have to figure out how to mitigate those failures.”
So they are working constantly at figuring out how to monitor what they have and figure out what the equipment is doing, and how they can best optimize the performance. We talked earlier about battery runtimes, for example. Sometimes they might get short or sometimes they might be long.
As these companies get into this step and repeat function, they are going to get consistency of their methods of procedure. They’re going to get consistency of how their operations teams run their physical infrastructure. They are going to think about running their equipment in ways that is nontraditional today but will become the norm in the next generation of data centers. And then they are going to look at us and say, “Okay, now that I have normalized my design, can I use rapid deployment configuration? Can I put it on a skid, in a container? Can I drop it in place as the complete data center?”
Well, we build it one piece of equipment at a time and stitch it all together. The question that you asked about monitoring, it’s interesting because we talked to a major company just last month. Steve and I were visiting them at their site. And they said, “You know what? We spend an awful lot of time figuring out how our building management system and our data exchange happens at the site. Could Vertiv do some of that in the factory? Could you configure our data acquisition systems? Could you test them there in the factory? Could we know that when the stuff shows up on site that it’s doing the things that it’s supposed to be doing instead of us playing hunt and peck to figure out what the issues are?”
We said, “Of course.” So we are adding that capability now into our factory testing environment. What we see is a move up the evolutionary scale. Instead of buying separate boxes, we are seeing them buying solutions — and those solutions include both monitoring and controls.
Steve didn’t even get a chance to mention the industry-leading Vertiv Liebert® iCOM™ control for thermal. These controls and monitoring systems allow them to increase their utilization rates because they know what they have and what it’s doing.
Gardner: It certainly seems to me, with all that we have said today, that the data center status quo just can’t stand. Change and improvement is inevitable. Let’s close out with your thoughts on why people shouldn’t be standing still; why it’s just not acceptable.
Innovation is inevitable
Madara: At the end of the day, the IT world is changing rapidly every day. Whether in the cloud or down at the edge, the IT world needs to adjust to those needs. They need to be able to be cut out enough of the cost structure. There is always a demand to drive cost down.
If we don’t change with the world around us, if we don’t meet the requirements of our customers, things aren’t going to work out – and somebody else is going to take it and go for it.
Panfil: Remember, it’s not the big that eats the small, it’s the fast that eats the slow.
Madara: Yes, right.
Panfil: And so, what I have been telling folks is, you got to go. The technology is there. The technology is there for you to cut your cost, improve your speed, and increase utilization. Let’s do it. Otherwise, somebody else is going to do it for you.