White Paper

7
Availability
Both AC and DC power systems can be
designed to achieve high availability. The
highest availability facilities will likely con-
tinue to use 2N redundancy because of its
ability to support a dual-bus architecture
that can eliminate downtime from system
failures across the entire power distribution
chain. However, that availability comes
with a cost, both in terms of upfront
equipment and ongoing operating
efficiencies. Many organizations can
achieve desired levels of availability from
N +1 redundancy in which redundancy is
maintained at the module level.
At the system level, there are two ways that
DC power systems offer high availability.
First, the DC power conversion system has
fewer components than a comparable AC
system, which contributes to a higher
mean time between failure (MTBF) rate and
more uptime. In addition, the DC UPS uses
an array of discrete rectifiers to deliver con-
ditioned, isolated power to a distribution
bus. These rectifiers provide built-in redun-
dancy; the system can accommodate the
failure of any individual rectifier without
immediately affecting performance or
capacity. Individual rectifiers can be safely
hot-swapped out in the field without
impacting system operation, thus minimiz-
ing system mean-time-to-repair (MTTR),
a major contributor to unavailability.
Scalability
With equipment and rack density rising
steadily, the power system can become a
constraint to growth. As a result, modular
approaches to data center design and
expansion are gaining in popularity as
demonstrated by the trend toward
“containerized” data centers.
In the white paper “Phase Balancing:
The Last Few Inches of a High-
Efficiency Power System”
4
, Server
Technology analyzes the impact of
unbalanced loads on a 30A, 240/415V
3-phase circuit loaded in a Wye
configuration. If the load is balanced,
the current through each input phase
is 8A, and losses can be calculated at
19 watts per 100 feet of cable. In
the most unbalanced case, current
through one phase is 24A and losses
escalate to 115 watts per 100 feet of
cable. This shows that power loss and
heat generated in feeder cables can
increase by as much as 600 percent
because of unbalanced loads.
Additionally, keeping track of which
base loads are on which phase can be
time consuming and tedious depend-
ing on the sophistication of the site
and the tools available. Load balancing
issues are also specific to three-phase
AC power and are not a concern
with DC.
Power-Related Heat Loads
After the IT equipment itself, the cool-
ing system is the next largest user of
power in the data center. Not only is
cooling required for the heat gener-
ated from the IT equipment, but it
also must offset the heat generated
by power conversion losses. With
fewer required conversions and
greater overall efficiencies, a DC
power system generates less heat
than an AC system, reducing data
center cooling energy consumption.
As a rule of thumb, each watt of heat
generation removed from the data
center leads to an additional 1.4 to 2
watts saved in cooling.
With a complete data center power and
cooling infrastructure integrated into a fully
enclosed container, these systems can be
delivered to a site and quickly plugged into
power, communications and chilled water
systems to enable additional computing
capacity. With its built-in redundancy and
compact footprint, a DC UPS is an excellent
solution for these applications.
This same philosophy is also manifesting
itself in the move toward row-based equip-
ment, which enables modular expansion
of existing facilities or support for high
density rows. Both AC and DC UPS systems
provide scalability in a row-based format,
but the DC UPS has the benefit of built-in
redundancy and will occupy at least 50
percent less floor space than a comparable
row-based AC system.
Cost
The main purpose of the critical power
system is to eliminate power-related
downtime; any cost comparison should
consider the level of availability required
and the cost of downtime. In general, row-
based DC power applications will be less
expensive to install, operate and maintain
than a comparable AC system and will
support redundancy levels of at least N + 1.
Some savings may be offset by increased IT
equipment costs in a DC system—list prices
for servers with DC power supplies can be
up to 10 percent higher than the more
common AC servers. These costs may be
negotiable and should come down with
the higher volumes that accompany
increased adoption. Overall, a DC-based
end-to-end power architecture can be as
much as 30 percent less expensive than
an AC system, depending on many of the
factors above.