Comparing Data Center Power Distribution Architectures

Executive summary:

Significant improvements in efficiency, power density, power monitoring, and reconfigurability have been achieved in data center power distribution, increasing the options available for data centers. This research compares five power distribution approaches including panelboard distribution, field-wired PDU distribution, factory-configured PDU distribution, floor-mount modular power distribution, and modular busway, and describes their advantages and disadvantages. Guidance is provided on selecting the best approach for specific applications and constraints.

Revision notice:

This original version of this research focused on the benefits of floor-mount modular distribution. This revision discusses the five common distribution approaches and provides guidance for selecting the best approach for a given application.

Introduction:

Many existing data centers utilize the same power distribution architecture to get power to their IT equipment that was developed for data centers approximately 40 years ago. There have been dramatic changes, however, in how power is utilized in data centers that have challenged this architecture, specifically driven by increasing power density, the increasing number of separate IT devices within the data center, and the need to add and remove IT devices on a continuous basis. Improved systems allow IT racks to be installed or changed without any new wiring, distribute power overhead, support rack densities up to 30 kW with a single flexible power feed, improve electrical efficiency, are instrumented for power at the branch circuit, and have a standard capacity management system.

This research compares the five power distribution approaches seen in data centers today, including panelboard distribution, traditional field-wired PDUs (power distribution units), traditional factory-configured PDUs, floor-mount modular power distribution, and modular overhead (or underfloor) power busway (see Figure 1). The advantages and disadvantages of each approach are discussed, and guidance is provided on selecting the optimal approach for specific applications and constraints.