Data Center Journal

VOLUME 51 | AUGUST 2017

Issue link: https://cp.revolio.com/i/860140

Contents of this Issue

Navigation

Page 22 of 32

20 | THE DATA CENTER JOURNAL www.datacenterjournal.com industry cHAnges D ata centers are generally growing in computing and storage capac- ity. e move from enterprise to colocation in the 2012–2016 period led to the construction of massive data centers. en, the rapid growth of hyperscale data centers from 2015 and 2016 is leading to the further "industrializa- tion" of these facilities: massive scale, in-house innovation in architectures and equipment, and large networks of data centers that offer new ways of providing redundancy. To improve the economics of data centers and to cope with the demand for computing and storage capacity, the equipment density in kilowatts per rack has jumped from 3–8 kW a few years ago to 20–35 kW today. Data center operators are going global. Colocation providers have been globalizing their operations to serve their clients—and to grow their businesses. Hyperscale data centers have been expanding globally to serve their internal customers as well as us- ers of their cloud-based applications. Customers from all over the world are influencing design choices, including distribution voltages, when they take space in colos far from their home countries. And European data-priva- cy laws that become effective in May 2018 are contributing to the expan- sion of U.S. data center companies in Europe. Imperatives for Higher Efficiency and Productivity Although the need for data center ca- pacity is growing rapidly, the industry has matured during the 2012-17 pe- riod. ere is more competition, and colos are seeing hyperscale companies as both customers and competitors. is situation is driving a push for productivity, which in turn is driv- ing standardization of architectures, equipment and processes. It's also driving open standards to optimize the footprint, energy consumption and cost of equipment. Edge Data Centers Applications that require very low la- tency and that involve heavy two-way traffic between the user and the data source are triggering construction of more edge data centers. ese facili- ties are physically close to the data users. ey are generally smaller than typical data centers and are economi- cally viable only if they are highly automated—that is, their operation follows a "lights-out" approach. Technology Soware has enabled ever more virtu- alization and geodiversity to maxi- mize asset efficiency and provide new ways of implementing redundancy. In addition, sensors have become smaller and more cost effective. So- ware and tools to store and analyze data from sensors, as well as make de- cisions using that data, are proliferat- ing. Direct-current (DC) distribution has seen deployment in a few places, but it has yet to see wide commercial- ization. Because some innovations have occurred in architectures based on alternating-current (AC) distribu- tion and localized DC distribution, it's unclear that DC distribution will ever garner adoption a larger scale. Regulations OSHA is driving the industry to achieve ever higher safety levels. e National Fire Protection Association's NFPA 70E code is receiving an update for 2018. ere is a heightened sensi- tivity and awareness of safety-related guidelines among both operators and their personnel. Cost of Outages Several highly visible and expensive outages occurred during 2016 and 2017. One notable outage caused a nine-figure fallout and massive business disruption. Experts have pointed to a few primary failure modes where personnel errors—not equipment failures—are the primary cause. Although many of these errors could have been prevented through better SOPs (standard operating procedures) and training, others will require a rigorous review of processes that cover preventative maintenance, power architecture and cybersecurity (the fastest-growing threat). tHe future of power distriBution in dAtA centers e two main types of data cen- ters will likely become more distinct over time. e first type is hyperscale data centers, which can afford to em- ploy engineers who design and build unique architectures and equipment in partnership with OEMs. e other type is smaller data centers, which are generally better off adapting the products that power-distribution OEMs offer. Architectures Hyperscale data centers are already using nonstandard designs and equipment to eliminate AC-DC-AC conversions, reducing cost; provid- ing redundancy that takes advantage of their multiple data centers; and distributing power at higher voltages. e architectures that some of these data centers have tested will evolve into designs that other large data centers will prefer.

Articles in this issue

Links on this page

Archives of this issue

view archives of Data Center Journal - VOLUME 51 | AUGUST 2017