Evaluating the Opportunity for DC Power
Mark Murrill is program manager of Data Center DC Power for Emerson Network Power.MARK MURRILL
Emerson Network Power
With data center managers struggling to increase efficiency while maintaining or improving availability, every system in the data center is being evaluated in terms of its impact on these two critical requirements. The power system has proven to be one of the more difficult systems to optimize because efficiency and availability are often in conflict; the most efficient approach to critical power is rarely the most reliable.
One solution to power system optimization that deserves serious consideration is DC power. Since utility AC power must ultimately be converted to DC power for use by all silicon chip-based IT equipment and because stored energy systems (batteries, flywheel, etc.) provide DC power for backup, a DC power architecture requires fewer total conversions from grid to chip, creating the opportunity to reduce costs and increase efficiency. It also eliminates the need to de-rate usable capacity due to unbalanced loads, eliminating the concept of stranded power and allowing full utilization of power infrastructure.
The first decade of the twenty-first century was one of incredible growth and change for data centers. The demand for computing and storage capacity exploded, and many IT organizations struggled to deploy servers fast enough to meet the needs of their businesses. At the same time, trends to consolidate data centers and centralize computing resources resulted in the deployment of more powerful servers which created a dramatic rise in data center power consumption and density, leading to a 400 to 1,000 percent increase in rack density. This resulted in fewer opportunities for planned downtime while also increasing the cost of unplanned outages.
The dramatic increase in data center energy consumption has created both financial and regulatory challenges. Energy costs, which once had been relatively inconsequential to overall IT management, became more significant as the rise in consumption was exacerbated by a steady—and in some years significant— increase in the cost of electricity. As governments and utilities aim to avoid power shortages and control demand growth, high energy users like data centers have become more aware and are working to reduce any excess consumption with a new focus on energy efficiency.
The industry is now implementing server virtualization, higher efficiency server power supplies, and new approaches to cooling. Yet, while significant progress has been made in some areas, the critical power protection system has yet to be fully optimized in most cases. While individual components have been improved, the overall system complexity is high, which can create inefficiency and add operational risk. Faced with the choice of increasing system efficiency or adding risk, many continue to choose conventional approaches that deliver high availability but may not offer the highest efficiency.
A close examination of the available options reveals that, in many cases, efficiency can be improved without sacrificing overall availability. New, row-based DC power systems represent a practical and affordable solution for reducing data center complexity, increasing efficiency and enabling growth. Several reasons to consider row-based, 48V DC power in your next data center facility follow.
Exact end-to-end power system efficiency can be difficult and time consuming to accurately calculate due to the complexity of variables involved, but it is still useful to consider and approximate based on a given set of site conditions and priorities. In several published papers from the Green Grid and Intel, overall DC power architectures are shown to be 5-20% more efficient depending on the assumptions used. Factors that can influence these comparisons include: the presence of 2N, parallel bus redundancy; the impact of harmonics and load swings; availability of an energy optimization mode for near-peak efficiency at all load conditions; and the additional power required to cool waste heat generated by inefficient systems. In cases where these factors are present, the gaps between DC and AC widen.
A row-based DC power system offers high availability in two ways. Not only does it contain fewer components than a comparable AC system, helping to realize a higher mean time between failure (MTBF) rate and more uptime, but also uses an array of discrete rectifiers to deliver conditioned, isolated power to an internal distribution bus. This group of power conversion units (PCUs) provides built-in redundancy – the system can accommodate the failure of any individual unit without immediately affecting system operation. These units can be safely hot-swapped in the field which minimizes the system mean-time-to-repair (MTTR), a major contributor to unavailability.
With equipment and rack density steadily rising, the power system may become a constraint to growth. As a result, modular approaches to data center design and expansion are gaining in popularity. Whether this takes the form of a pre-fabricated structure or just a defined, standard layout, row-based DC power is ideal for these applications enabling cost efficient speed to market and operational deployment advantages.
The main purpose of the critical power system is to eliminate power-related downtime; any cost comparison should consider the level of availability to be achieved and the cost of downtime. In general, row-based DC power applications will be less expensive to install, operate, expand and maintain than a comparable AC system while supporting redundancy levels of at least N + 1.
Direct current is already a fundamental part of your IT infrastructure: critical loads consume DC power, renewable sources generate it, and batteries store it; we know that the electric grid distributes AC power. The question becomes, where is the optimal point at which to convert AC to DC power while providing suitable protection from outages? If the conversion occurs too early, DC power must be transported long distances, which requires large conductors to reduce losses. If it occurs too late, additional conversions are introduced into the process which compromise efficiency and reliability and increase costs. Therefore, in many data center applications, the ideal point for power conversion and energy storage is at the row; a row-based, 48V DC UPS allows for optimal efficiency, availability and flexibility in these applications.
Industry Perspectives is a content channel at Data Center Knowledge highlighting thought leadership in the data center arena. See our guidelines and submission process for information on participating. View previously published Industry Perspectives in our Knowledge Library.
Thanks for the support. we identified these value propositions many years ago and your support for our design is a long time coming. all of what you state is a fact and is a fact in all DC systems at any voltage. Neg 48 VDC is a good high voltage bus for servers but less desirable for distribution. Validus has operating designs 48/380/575 row or rack based. You should give a call we would working together would be beneficial. Thanks again, good article. Rudy Kraus CEO VDC
Rudy, thanks for your comment. No question the values and benefits of DC power have been around a long time, especially at -48V as deployed in telecom applications; Emerson (and predecessor companies) has been a leader in this market for decades. As you point out though, a potential drawback to 48V is that large conductors are required for distribution over long distances which can add to installation cost. Putting the power conversion in-the-row and integrating battery backup for short duration, as with our new 48V DC UPS, allows for much shorter distribution runs. This dramatically reduces the total copper required and makes these deployments very economical at today’s rack densities.
Watch for upcoming news on study and comparison of DCV vs. ACV at the rack that has been done at Duke Energy and sponsored by EPRI. Claims 15% advantage of DC over AC!