28
Data centers are specialized critical facilities designed to
house dense collections of computing equipment: servers,
storage arrays, switches, routers, and related items. Data
centers range from small room-sized facilities with a total
electric load of some tens of kilowatts to very large,
purpose-built facilities and campuses dissipating tens of
megawatts. Conceptual designs and initial planning for
100 MW data centers are in progress.
Data centers typically incorporate a substantial collection
of electric power and cooling equipment. Nearly all data
centers are designed to supply continuous power and
cooling
1
and incorporate combinations of redundant
assets, standby equipment, and energy storage to achieve
this goal. A typical large data center includes:
• Redundant electrical distribution systems supplying
“dual cord” loads
• Dual cord computing equipment; so long as power is
available on either cord, the equipment will operate.
Normally power is provided to both cords.
• Standby diesel engine/generators and associated fuel
stores
• Uninterruptible Power Supplies (UPS) and associated
kinetic or electrochemical energy storage. UPS bridge
the gap between loss of utility power and availability of
power from standby generators.
• Redundant computer room air handlers
• Redundant water cooling towers and fans
• Looped water distribution piping designed to allow
isolation and repair of damaged components while
operating
• Redundant refrigeration assets (some modern data
centers reduce or eliminate mechanical refrigeration)
The purpose of these complex, highly redundant systems
is to increase the reliability of the data center. The owners
and users of the facility expect it to operate even if the
local electric utility experiences an outage, and during
extremes of hot or cold weather.
PRESENT DATA CENTER METRICS
Popular data center metrics include Power Utilization
Effectiveness (PUE) and the Tier Classification system, a
proprietary rating system developed and administered by
the Uptime Institute in the early 1990s.
PUE is related to efficiency; it is the ratio of total data
center electric load to the power consumed by computer
equipment. Power to drive refrigeration units, fans,
pumps, lights, and other non-computer loads increase
PUE. A smaller PUE is considered more desirable; a PUE
of 1 would seem to be the lower limit, but by reclaiming
heat rejected by computer loads it is theoretically possible
to operate facilities with PUEs less than 1.
The Tier classification system is a proprietary scoring
method administered by the Uptime Institute and certain
authorized parties. The Uptime Institute charges a fee to
award a “Tier rating” to any facility.
The Tier system is primarily a measure of system
redundancy, on a scale of Tier 1 to Tier 4. Tier 1 systems
have limited redundancy but typically incorporate a UPS
with a bypass path and a standby generator. Tier 2
systems incorporate redundant components, for example
multiple UPS modules, chillers, and standby generators. If
N units of each asset are required to serve the facility
load, N+1 or more are provided.
Tier 3 facilities incorporate dual power and cooling
pathways, and Tier 4 facilities provide still higher levels of
redundancy.
While the Uptime Institute claims that Tier levels are
associated with availability, analysis by MTech and others
2
,
3
have demonstrated that this is not the case. The claimed
availability levels for each Tier are the result of the
assumptions built into the system by the Uptime Institute
4
.
Most notably, both independent analyses conclude that
there is very little difference in availability between Tier 3
and Tier 4 facilities.
Even if the Tier system were an accurate predictor of
availability, it does not address reliability.
7X24 MAGAZINE SPRING 2015
by
Steve Fairfax