How the Internet of Things affects the data center

An enormous amount of processing and storage capacity is needed to keep pace with the Internet of Things.

By Bill Kosik, PE, CEM, BEMP, LEED AP BD+C, HP Data Center Facilities Consulting March 4, 2015

Learning objectives

  • Learn how the Internet of Things affects engineers.
  • Understand how the connection of devices will affect data centers.
  • Know how the Internet of Things will affect energy requirements.

The Internet of Things (IoT): an intriguing yet enigmatic term that affects roughly 2 billion people who live their day-to-day lives in some degree of dependence on the Internet. What is IoT? Simply put, IoT is the trend of connecting physical objects to the digital world. The term has been used increasingly in the last few years but was actually coined in 1999-still in the early stages of the Internet. Today, the information technology industry is in a constant state of change, shaping and being shaped by the needs of business,research, and private citizens (just to name a few). The most visible examples are the devices we use in everyday life, such as mobile phones, computers, smart watches, home security systems, automobile sensors, and the list goes on. These devices are not stand-alone. They connect to-and have bidirectional communications with-servers and storage devices located in some faraway data center.But here’s the clincher: according to Gartner, 25 billion devices will be connected to the IoT installed by 2020. This means a lot of processing and storage capability is needed, and needed in different ways and different times than what is currently practiced.

How is the IoT shaping our current and future lives? The use of social media, cloud storage, and mobile devices is the triumvirate that rules our daily lives. It is common to take a picture, upload it to a cloud storage site, and post it on a social media site. This is just a typical example of how we are connected to the digital world. But people also use their smartphones to pay bills, deposit checks, play music, scan a document, look up a movie review, pay for parking, order dinner, check out a library book, and gather key stats from a workout, to name a few. None of these things are particularly interesting or out of the ordinary. But the routineness of these activities drives home the point: As a society, we have become so dependent on our digital lives that we are craving for increasingly more of our current analog activities to be replaced by computer-based solutions. Consequently, our reliance on the IoT will continue to grow at a rapid pace (see Figure 1).

While this has significant implications on servers, storage appliances, and networking gear, it also impacts the data center cooling and power systems, specifically the systems’ scalability, capacity, and provisioning capabilities. To provide the required power and cooling to keep the IoT whirring along,building services must keep pace with the digital world; there is no choice but to comply. In fact,according to a National Research Council report, next-generation computing platforms will have a greater reliance on redundancy and fault tolerance as the rate of performance improvements begins to slow as Moore’s Law comes to an end in the next decade. And if done properly, with tight integration with the IT realm, there will be reductions (not simply efficiency gains) in overall data center energy use, while providing better computing ability. Power and cooling systems are at the "end of the pipe" while the IT systems are at the beginning. Any reduction in energy use to run servers, storage devices, and networking gear will cascade down and increase energy efficiency of power and cooling systems. This is the key message: The IoT will increase society’s access to the digital world and can reduce the energy required as compared to today’s standards.

Scalability

Based on improvements of IT hardware over the last decade, data center systems must adapt to operating at different workloads while maintaining energy efficiency and reliability. In electronics(including hardware, communication, and software), scalability is the ability of a system, network, or process to handle a growing amount of work in a capable manner or its ability to be enlarged to accommodate that growth. But scalability is a pretty slippery term. Not only can it be applied differently in multiple industries, but its meaning can vary within a particular industry. To confuse matters even more,we can talk about scaling "up" and scaling "out." For a business that relies heavily on computing capability, the demands of the business will outpace the ability of the computing system, potentially compromising the company’s mission. In this situation, scaling of the IT systems is necessary.But in the context of the IoT, how can we scale? Scaling "out"-in simple terms-means to add more nodes (computers, storage, etc.) to the system. Scaling "up"-also in simple terms-means to build up on a single node within the system, like expanding memory or adding more processors. These two concepts of scalability have different implications for power and cooling systems in the data center. Scaling outproduces an extensive power and cooling delivery requirement (more space, lower IT power density),while scaling up will produce an intensive requirement (less space, higher IT power density).

Capacity

Computers specifically designed to work in the IoT ecosystem must accommodate the new and diverse types of processing, networking, and storage requirements. And they might be very different from traditional rack-mounted servers or mainframes. They have a very high node density, which translates into high electrical density (W/sq ft). The high node count requires robust thermal management to deliver the proper temperatures inside the servers. Doing so often requires liquid cooling to counteract the heat dissipation resulting from the very demanding processor workload. For example, a major IT enterprise hardware manufacturer recently released a new style of server: Each 4.3U (in a standard server-rack configuration, one rack unit or 1U = 19 in. wide and 1.75 in. tall, which defines the minimum possible sizeof any piece of equipment) chassis accommodates up to 180 server cartridges that work off of common power, cooling, and communications components. A cartridge can be taken out of service (hot swap)without having to power down any components or disconnect network cables. This capability greatly improves system reliability and serviceability. And because each cartridge is "software defined," there can be different types of applications running on different servers, significantly boosting computing flexibility.Knowing that 10 chassis will fit in a standard server cabinet, up to 1,800 servers can be installed in one cabinet, yielding total power per cabinet of 80 kW. These types of servers are aimed at the enterprise audience (e.g., hosting, analytics, gaming, and telecom, to name a few) and can significantly reduce the number of traditional 1U, 2U, and blade servers.

With all of this computing power and the blurring of lines that divide hardware and software, it is easy to get caught with our head in the clouds (pun intended). Basic physics and thermodynamic laws cannot be evaded: the data center still must have the ability to power and cool the equipment it houses. Some of the most basic data center facility design, construction, assurance, and operation tenets still must be met:provide adequate physical space, consider both current state and future growth scenarios, supply necessary power for IT and cooling equipment, ensure that the ability to generate the air/water temperatures required by the IT equipment is achievable, and continually optimize energy use and operating costs.

Physical space:

There are many factors that drive the size of a data center facility. But when considering the IoT impact on the size of the facility, a few important ideas emerge. The use of high-density, software-defined servers can significantly reduce the amount of floor space required.Comparing a traditional cabinet-mounted half-width server installation requiring 10 cabinets of 800 servers to a system of software-defined servers requiring 6 cabinets with 6,400 servers, the overall data center floor space for the software-defined server system is approximately 50% less than the traditional option. This includes the IT equipment space and the power and cooling
infrastructure space. Looking at it from a different angle, on an area basis, this server system can deliver almost four times the computing power in the same amount of space as a traditional cabinet-mounted server system (see Figures 2 and 3, and Table 1).

Power for IT and cooling:

When there is a limitation on power availability, such as electrical utility capabilities or when installing IT gear into an existing data center, the IT equipment performance per Watt becomes an important parameter to maximize the computing power. A great example comes from the high-performance computing (HPC) sector. Across the sector, HPC machines will be used for many diverse applications, are located across the globe, and have hundreds of different hardware and power/cooling infrastructure combinations. In the HPC industry, there is a targeted effort to maximize the computer’s performance using the smallest amount of facility power.

Proper temperatures for IT equipment:

Reducing data center space and optimizing the computing performance per power use will become of no use if the IT equipment fails to operate properly due to temperatures that go beyond the envelope of permissible internal component limits. To combat this serious problem, many high-density server systems-including high-performance, software-defined systems-use internal liquid cooling. While liquid-cooled computers have been in operation for decades,they typically have been very static, monolithic mainframes that were used for very specific purposes.Today’s water-cooled computers are akin to cabinet-mounted servers that are hot-swappable without loss of cooling or power to the rest of the servers in the cabinet. This type of computer architecture greatly improves the cooling capability because coolant is supplied directly to the processors, memory modules,and graphics processing units. Depending on the type of computer, provisions for air cooling must be made because a portion of the heat generated by the processors and their support circuits will still be discharged to the air in the data center. For a large system, the heat discharged to the air could exceed 10 kW, which is equivalent to that of a moderately dense air-cooled server cabinet.

Energy cost reduction:

If cloud computing was a country, in terms of energy consumption,it would rank fifth in the world. From an energy engineering perspective, the best part of a computer system that is part of the IoT is its reduction in energy use (and cost) that comes along for the ride. There will be inherent cooling and heating energy reduction in reducing the overall size of the building. But the greatest energy savings comes when water-cooled IT equipment is used. By using water cooling, almost all of the fan energy associated with the air handling units (AHUs) that typically cool the data center is eliminated. And when water is used as a cooling medium sent directly to the internal components of the computer, the water temperature can be as high as 80 to 100 F. Using water temperatures like these means that in most parts of the world, all cooling can be done using total free cooling or free cooling using only minimally mechanical cooling. This creates a huge possibility to have incredibly powerful data centers that fuel the IoT using only a fraction of the power that is used in a traditional data center.

To illustrate this, side-by-side energy analyses were developed using two geographic locations (Sao Paulo, Brazil, and Amsterdam, Netherlands) and two types of data centers, one using air-cooled enterprise servers and the other using an HPC, water-cooled system. The purpose of the analysis is to understand how the two different server types affect the energy use of the cooling systems. In these analyses, the base cooling systems are water-cooled chillers and AHUs. Key findings of these analyses include:

  • The primary energy savings strategy for the HPC system, from a cooling perspective, is being able to use very warm water to keep the systems at the proper temperatures.
  • Using water cooling works well in a very warm climate, such as Sao Paulo, when compared to air-cooled servers.
  • The differences in energy for cooling in Sao Paulo are substantial: The water-cooled system consumes more than 50% less energy than the air-cooled server systems (see Table 2).
  • In a more moderate climate, such as Amsterdam, the energy to cool the water-cooled systems is 37% less than the equivalent air-cooled server systems (see Table 3).
  • In more moderate climates, the water-cooled chiller plant is able to use economization more hours per year compared to the very warm climate. This is why the percent of energy reduction for the water-cooled system is less than the very warm climate.
  • When using water-cooled equipment, the energy use for data center fans is reduced by a large amount, going to zero depending on the server’s thermal management tactics. Certainly, fans are still required for other areas of the data center, such as UPS rooms and other infrastructure spaces.
  • Depending on the location and the operational goals of the HPC systems, mechanical cooling can be completely eliminated. Large central plant equipment, such as chillers and pumps, can be eliminated completely and replaced by much smaller "booster" chillers designed to inject cold water into the cooling loop when necessary to keep the loop at the required temperature. The rest of the time, heat rejection equipment, such as cooling towers or closed-loop coolers, is used to meet the requirements.

Provisioning

As a part of a broader facility-as-a-service strategy, power and cooling systems will be operated using a greater degree of software automation, tightly linked to the actual workload on the computers. Using this approach on HPC systems can reduce the energy required to run the massive computational workloads for which these systems are famous. It also can increase node reliability and reduce backbone traffic by distributing those data across different racks. Having this distributed approach in place will theoretically improve energy use because nodes that have duplicated data can go into a low power state or even shutdown entirely based on how the workload is instructed to move through the system. And this also increases reliability if a server or rack is lost due to some type of failure-the job just moves on to a node that has the duplicated data and continues without missing a beat. This type of workload shifting requires agile power and cooling systems that can respond to data that are directing where the workload will be(either in the data center or over multiple data centers) and respond accordingly. These requirements come from the increased expectations of what is required of enterprise business: more reliance on mobile apps, access to critical documents such as in a health care setting, organization and storage of material that must be saved and retrieved for a long time, and big data workloads that require agile processes to perform complex and cohesive data analysis.

What’s next?

Based on industry leaders, analysts, and corporate executives responsible for charting the course of multibillion-dollar organizations, the IoT-driven by IT systems and facilities that make it possible-will continue to grow in number, but also in sophistication. For example, considerable research has already been conducted on autonomous cars "that cannot crash." The success of these cars relies heavily on computation and data processing done in data centers, remotely connected via the Internet. Another example (on a much smaller scale) is personal health monitoring. Far beyond today’s wearable fitness devices, microelectromechanical systems, sensors, and low-power radios embedded in a variety of devices will generate a wealth of information on one’s personal activity pattern, providing advice on howto alter behavior to improve health conditions. Although the raw data are gathered locally, the heavy lifting of data aggregation and analysis will occur in a data center somewhere, linked by the Internet. Based on these examples (and dozens of other cutting-edge technologies currently being developed by private industry, universities, and government agencies), the IoT will continue to drive the need for more reliable and efficient IT systems and data centers.

Another technology is being developed: Based on the daily intake of information gleaned from Internet,"augmented cognition" will help us sort through the digital clutter and identify the truly important action items. It is ironic that a solution to de-clutter our digital world comes from the same source that created the mess in the first place. This is the future.


Bill Kosik, PE, CEM, BEMP, LEED AP BD+C, is a Distinguished Technologist at HP Data Center Facilities Consulting (DCFC). He is responsible for execution of project work, training and mentoring of internal engineering and consulting teams, research and analysis of topics related to data center energy use, and industry presentations and writing assignments. He is one of the main technical contributors who have shaped the DCFC technical expertise in energy use optimization and sustainable design practices in data centers. He is a member of the Consulting-Specifying Engineer Editorial Advisory Board.