By Marc Cram, Director of New Market Development, Legrand
As data centers address more expansive and exceptional challenges, their power distribution equipment must meet those performance needs. Server cabinets and racks, even individual servers, need to be designed for maximum adaptability to the growing power consumption requirements of their unique and demanding environments.
Whether dedicated to supercomputing or artificial intelligence, data centers are unique in form factor and physical architecture. Sometimes they’ll fit into an existing building on campus, with a retrofit of new infrastructure to support the additional demands placed on the power and cooling systems of the facility. Now and then, a new facility is designed and built to house the data center. Administrators must find custom solutions for delivering power, cooling, networking, and so forth in both instances.
Edge computing is designed to put applications and data closer to devices—and their users. But it brings a different set of challenges than the massive data centers used in supercomputing and AI applications. In many cases, space is significantly limited, with less room for the power distribution equipment. Because edge computing takes place remotely, you need to maintain remote connectivity and, at times, rectify any issues while being in a different location.
Big Data Requires Big Power
Whether streaming “Stranger Things” on Netflix or posting an mp4 file on TikTok, both cause a domino effect that leads to power consumption, not just with personal devices but in data centers worldwide where that information resides.
Then throw in high-performance computing (HPC) installations tasked with carrying out large-scale computations to solve complex problems that require a lot of data or have enormous computing power at their disposal, and you have added significantly to global power consumption.
In 2020, some of the world’s largest data centers contained tens of thousands of IT devices and required more than 100 megawatts (MW) of power to run, equivalent to powering around 80,000 US households (US DOE 2020).
With this massive demand for power comes the challenge of managing power distribution on a more granular level.
Individual Applications Create Unique Challenges
Off-the-shelf and semi-custom solutions meet the requirements of most data center applications. However, the need for continuing improvements in efficiency and sustainability prompts many HPC installations, AI applications, hyperscale data centers, and telecom operators to seek singular solutions to layout, power density, cooling, and connectivity.
While supercomputing needs everything physically close together to maximize throughput, AI wants to be on specialized processors, and by its very nature, edge computing is fundamentally distributed. Therefore, each application brings unique challenges for power distribution.
With supercomputers, the space dedicated to processing and CRACs (computer room air conditioners) leaves little room for distributing power going into the racks. A situation like this has potential challenges for deploying Power Distribution Units (PDUs), necessitating a customized solution. There may be little or no room at the back of the rack for a zero U PDU, requiring the PDU to sit on the racks’ sides. At the same time, taller racks with more servers generate high outlet density situations. And the need for high power density for the racks may be addressed through intelligent rack PDUs with monitoring capabilities.
Artificial intelligence requires astonishing amounts of computing power and electricity to run training algorithms. A classic aspect of AI applications is their high internal bandwidth between boxes/nodes and optical interconnections.
When designing a power distribution plan for an AI facility, you may need a PDU to help with capacity planning and maximizing electrical power utilization. Because high density and higher power installations test the limitations of standard PDUs, purpose-driven, custom-designed PDUs may be needed.
Edge data centers are smaller computational devices closer to the populations they serve. They deliver cloud computing services and cached content to end users. An edge data center also reduces operational costs, thanks to the reduced bandwidth requirement and low latency.
However, edge data centers’ environments often require them to function across broad operating temperature ranges, which may necessitate using environmental sensors to safeguard against temperature and power extremes outside the operational capabilities of the equipment. Additionally, edge computing is an ideal case for remotely monitoring power consumption.
Because edge data center environments are remote, a different power distribution architecture is needed, one that requires:
- PDUs that have onboard communications capable of scheduling outlet power on and off
- PDUs capable of shedding the power load to maximize battery power uptime if the unit exceeds thresholds
- An operating environment which dictates that the PDUs go beyond the usual 0-60 degrees Celsius
The Value of Custom PDUs
Off-the-shelf and semi-custom solutions for remote access, power, and white space infrastructure meet the needs of conventional data center applications. However, supercomputing, AI, and edge data centers often require customized solutions. Part of the customization must include careful thought to power distribution throughout the rack and, just as important, monitoring the rack’s environmental conditions. Highly specialized data centers deserve PDUs that meet their rack’s unique power and monitoring requirements. In these cases, it’s best to work with a trusted partner to design and build a PDU that can fit the needs of your particular applications today, tomorrow, and years from now.
Marc Cram is Director of New Market Development for Legrand’s Data, Power, and Control division, which includes the Raritan and Server Technology brands. A technology evangelist, he is driven by a passion for delivering a positive power experience for the data center owner/operator. He earned a bachelor’s degree in electrical engineering from Rice University and has more than 30 years of experience in electronics.