Cisco Cisco UCS 6120XP 20-Port Fabric Interconnect Libro bianco
White Paper
© 2011 Cisco and/or its affiliates. All rights reserved. This document is Cisco Public Information.
Page 1 of 13
Power Management in the Cisco Unified Computing System:
An Integrated Approach
An Integrated Approach
What You Will Learn
During the past decade, power and cooling have gone from being afterthoughts to core concerns in data center
construction and operation. Industry participants have improved the efficiency of almost all the elements in the power
and cooling chain, from individual chip efficiency, to system design, to efficient data center cooling. Currently, users
can buy the most efficient server designs ever available in terms of throughput per watt, and the macro design of
new data centers is vastly improved over those of even a decade ago.
The weakest link in the system remains the management of power at the system and rack and row levels, where
much of the power consumption occurs. While system vendors can manage power within a single chassis, until now
no vendor has provided a comprehensive and usable solution to rack- and row-level power management, especially
for cases in which flexible management of power peaks is required. The Cisco Unified Computing System
™
is the
first solution to meet this challenge.
The Cisco Unified Computing System provides power management across groups of Cisco
®
UCS chassis and racks
that can enable throughput improvement of up to 30 percent for a traditional data center at no additional capital cost.
This capability is integrated into the Cisco UCS architecture.
This document explores the following topics:
●
What is power capping and why is it a good idea?
●
Why is power capping rarely used even though it has been available for years in limited implementations?
●
How does the Cisco Unified Computing System solve the power management problem?
Multiple Places to Save Power: Not All Are Equal
Before looking in depth at power management and capping, you should understand the context within which Cisco
has been working. The fundamental goal of reducing power consumption per workload unit has been the subject of
multiple-year and multi-vendor effort, with innovations occurring at multiple points within the data center as well as
within the computer systems themselves. While some elements of the power and cooling chain are outside the
current purview of a system vendor such as Cisco, within the bounds of the system there are multiple opportunities
for power and cooling efficiencies.
1
Figure 1 shows the technologies that the Cisco Unified Computing System has
implemented to reduce power consumption at the server and chassis levels.
2
1
This document uses the terms “power” and “power and cooling” somewhat interchangeably in discussions of the overall problem,
since cooling is additive to and directly correlated with power consumptions. When the document uses “power” in its literal sense to
refer solely to electrical power, the meaning will be clear from the context.
refer solely to electrical power, the meaning will be clear from the context.
2
The power consumption challenge is even more complicated than what is shown here. The server exists within a larger context
that includes power distribution and the data center’s power and cooling architecture. From best to worst cases, these extrinsic (to
the server) factors can make almost a 2 to 1 difference in overall power utilization efficiency (PUE), a metric of data center power
efficiency.
the server) factors can make almost a 2 to 1 difference in overall power utilization efficiency (PUE), a metric of data center power
efficiency.