HP ProLiant Server Power Management

HP ProLiant Server Power Management
HP ProLiant Server Power Management
Red Hat Enterprise Linux 6.1
Technical white paper
Table of contents
Abstract ........................................................................................................................................ 2
Introduction ................................................................................................................................... 2
HP Power Regulator ........................................................................................................................ 2
HP Power Capping ......................................................................................................................... 5
Power monitoring with HP iLO 4........................................................................................................ 5
Power Capping Demonstration with HP iLO 4 ...................................................................................... 6
HP Insight Power Manager ............................................................................................................... 8
ProLiant Power Management with RHEL 6.1 ........................................................................................ 8
Collaborative Power Control with RHEL 6.1 ......................................................................................... 9
Idle Power States (C-States) with RHEL 6.1......................................................................................... 10
Additional RHEL 6.1 Power Management Features ............................................................................. 11
Summary..................................................................................................................................... 12
For more information ..................................................................................................................... 13
Call to action ............................................................................................................................... 14
Abstract
Power management is crucial to data center power provisioning. This document provides a brief
overview of the processor-based power saving features supported on HP ProLiant servers and the
power management features such as Power Regulator, Power Capping and Collaborative Power
Control that are embedded in the ProLiant platforms. This document also discusses how these features
are used and their relationship to the Red Hat Enterprise Linux (RHEL) 6.1 operating system, including
new features available with ProLiant Gen8 Intel-based servers and RHEL 6.1.
Introduction
The RHEL 6.1 operating system and the HP ProLiant servers together use processor-based features to
achieve better power efficiency for processors. The processor-based features include:
• Performance states (P-states) define a set of fixed operating frequencies and voltages, where P0
represents the highest operating frequency and voltage. You can save power by entering P-states
with lower frequency and voltage levels. Either the platform firmware or the operating system
controls the P-states.
• Power states (C-states), excluding the C0 state, represent idle states and determine the power
consumed when a processor is idle. C0 is a non-idle state with higher C-states representing idle
conditions with increasing power savings. The operating system controls the C-states.
• Throttle states (T-states) define a set of fixed frequency percentages which can be used to regulate
the power consumption and the thermal properties of the processor. ProLiant systems may reserve
the use of T-states for the system firmware.
In addition, ProLiant servers are also capable of utilizing the various processor states to support
innovative power management features that are operating system independent and are implemented
in the hardware and firmware:
• HP Power Regulator provides a facility to efficiently control processor power usage and
performance, either statically or dynamically depending on the mode selected.
• HP Power Capping allows an administrator to limit the power consumed by a server.
• HP Dynamic Power Capping has the additional feature of ensuring that the power limit set by an
administrator is maintained by reacting to a spike in server workload more rapidly than basic HP
Power Capping.
The Power Regulator and Power Capping technologies are designed to work in conjunction with each
other. To make the operating system aware of Power Capping, HP provides the Collaborative Power
Control technology. This is a two-way communication mechanism established between the operating
system and platform firmware, and can be used by the operating system and hardware
collaboratively to choose the appropriate performance level for the server. Support for this technology
is present in both RHEL 6.1 and on ProLiant Gen8 servers.
HP Power Regulator
HP Power Regulator is a configurable processor power usage feature which allows you to choose
from several options for the server to manage P-states or to delegate control of regulating P-states to
the operating system.
2
HP Power Regulator is implemented within the firmware on both Intel-based and AMD-based ProLiant
servers.1 ProLiant servers provide the following HP Power Regulator modes, that you can select from
the ROM Based Setup Utility (RBSU) or through HP iLO 4:
HP Dynamic
The firmware is capable of managing the P-states. However, when the
Collaborative Power Control (CPC) setting is enabled in RBSU, the OS and the
firmware collaborate to attain the desired frequency for a processor. When CPC
is disabled, this mode allows the firmware to exclusively control the P-states of a
processor to match the server load. On HP ProLiant Gen8 servers, HP Dynamic is
the default mode with the CPC setting enabled.
HP Static Low
The firmware controls the P-states. The P-state of the processor is static and it is set
to the P-state which corresponds to the lowest operating frequency supported by
the processor.
HP Static High
The firmware controls the P-states. The P-state of the processor is static and it is set
to P0 which corresponds to the highest operating frequency supported by the
processor.
OS Control
The RHEL 6.1 operating system controls the P-states and it manages the
P-states according to the policy set by the administrator via the OS.
For the HP Static Low and HP Static High modes above, you are advised to disable CPC to ensure
that the firmware has exclusive control of the P-states. CPC is located within the Advanced Power
Management Options in RBSU. This causes RHEL 6.1 to report in the /var/log/messages file and
in the dmesg output that CPU frequency scaling is not utilized on the server.
The OS Control mode allows the ProLiant platform firmware to delegate the duty of managing P-states
to the RHEL 6.1 operating system.
You can adjust the Power Regulator Settings through the RBSU or the HP iLO 4 interface as shown in
Figure 1. You must reboot the system to change the transitions to and from the OS Control mode but
you can change the system between the other three modes dynamically.
1
For detailed information on HP Power Regulator support across the different generations of ProLiant platforms, see:
http://h18013.www1.hp.com/products/servers/management/ilo/sup_servers.html
3
Figure 1: Configuring Power Regulator and Power Capping settings via iLO4
To adjust the CPC setting, you have to access RBSU as shown in Figure 2. Modifying this setting
requires a system reboot to take effect.
Figure 2: Configuring the Collaborative Power Control (CPC) setting via RBSU
For more information on HP Power Regulator technology, see:
http://h20000.www2.hp.com/bc/docs/support/SupportManual/c00300430/c00300430.pdf
4
HP Power Capping
Power Capping satisfies data center power provisioning requirements by allowing the data center
administrator to provide a power budget to a single-server or a group of servers. The ProLiant
platform enforces that limit by changing the processor P-states and T-states in an operating system
independent manner. Power Capping is independent of the HP Power Regulator setting and can
occur in any setting. When server power is being capped under OS Control mode, the firmware
overrides the power management instructions from the operating system for the duration of the
capping.
As shown in Figure 1, you can use HP iLO 4 to configure a power cap. HP iLO displays important
information about maximum available power for the power supply, the peak observed power, and
the minimum observed power for the server. With this information, you can select an appropriate
power cap, either by specifying the absolute maximum watts or a percentage of the maximum
observed power of the server.
For more information on Power Capping, see:
http://www.hp.com/go/dpc
For an in-depth presentation on the Power Capping technology, see:
http://h20000.www2.hp.com/bc/docs/support/SupportManual/c01549455/c01549455.pdf
Power monitoring with HP iLO 4
HP iLO 4 supports the facility to monitor current power consumption along historical timelines. As
shown in Figure 3, HP iLO 4 displays the current power consumption as well as the peak and
average power consumption for the past 24-hour and 20-minute time periods.
5
Figure 3: HP iLO 4 power meter readings for 24-hour and 20-minute time periods
For more information on the HP Integrated Lights-Out (iLO) management technology, see:
www.hp.com/go/iLO
Power Capping Demonstration with HP iLO 4
This section demonstrates the HP Power Capping functionality by increasing workload on a ProLiant
server under the RHEL 6.1 operating system. Figure 4 displays the iLO configuration setting for a
server where the capping threshold is set to 180 watts. This means the maximum power consumption
will be limited at approximately 180 watts. Figure 5 shows that when power capping is not set, the
power consumption of a server increases when there is an increase in workload. You can also see
that the maximum power was 227 watts while the minimum power was 209 watts in the past 5
minutes. The average power was 213 watts. Figure 6 demonstrates that when capping is enabled,
the platform limits the power consumption to 180 watts even with an increase in workload in order to
satisfy the power budget set by the user.
6
Figure 4: HP Power Capping threshold configuration
Figure 5: Power consumption without capping
7
Figure 6: Power consumption with capping
HP Insight Power Manager
HP Insight Power Manager (IPM) is an integrated power monitoring and management application that
provides centralized control of server power consumption and thermal output at the data center level.
With IPM, users can change the power cap settings on groups of servers at a time. IPM is a plug-in
for the HP Systems Insight Manager (SIM) that monitors and controls HP ProLiant and HP Integrity
servers throughout the data center.
For more information on IPM, see:
http://h18000.www1.hp.com/products/servers/management/ipm/index.html
For more information on SIM, see:
www.hp.com/go/sim
ProLiant Power Management with RHEL 6.1
RHEL 6.1 manages the power usage of ProLiant servers by adjusting the processor P-states when the
HP Power Regulator setting in RBSU is configured in OS Control mode. Typically within the Linux
operating system, a governor dictates the policy, while the actual P-state transition is accomplished by
a suitable P-state driver. RHEL 6.1 offers a choice of governors, each implementing a different policy
ranging from userspace, which enables the user space program (cpuspeed) to directly configure the
processor frequency, to performance, which selects the P-state corresponding to the highest supported
frequency. The default governor is the ondemand governor, which dynamically adjusts the processor
P-states to match the load on the server.
On Intel-based ProLiant platforms RHEL 6.1 natively supports the Intel Demand Based Switching with
Enhanced Intel SpeedStep ® Technology. On AMD-based ProLiant platforms, RHEL 6.1 supports
8
AMD's PowerNow! technology. The following table lists the P-state driver on Intel-based and
AMD-based ProLiant G6, G7 or later platforms under OS Control mode.
Processor Family
Intel ® Xeon ®
AMD Opteron ™
P-state driver
acpi-cpufreq
powernow-k8
In order for RHEL 6.1 to manage the power consumption of the processor, the firmware must
communicate information about the processor P-states and their associated frequencies to the OS.
You can find this information in the file and directories under /sys/devices/system/cpu.
Included in the RHEL 6.1 media is the cpufreq-util command (installed via the cpufrequtils007-5.el6.i686.rpm package) that provides information about the P-states of the processors in
the system in a user-friendly format. When used without arguments, cpufreq-info displays
information about all processor cores, including the P-state driver, the frequency range supported by
the processor, the available frequency steps (which are the P-states), the available and current
governors, and the current frequency. Example 1 shows how cpufreq-info also supports options
to display information specific to a CPU.
Example 1 Output for CPU 0 in OS Control mode
# cpufreq-info -c 0
cpufrequtils 007: cpufreq-info (C) Dominik Brodowski 2004-2009
Report errors and bugs to cpufreq@vger.kernel.org, please.
analyzing CPU 0:
driver: acpi-cpufreq
CPUs which run at the same hardware frequency: 0
CPUs which need to have their frequency coordinated by software: 0
maximum transition latency: 10.0 us.
hardware limits: 1.20 GHz - 2.00 GHz
available frequency steps: 2.00 GHz, 2.00 GHz, 1.90 GHz, 1.80 GHz,
1.70 GHz, 1.60 GHz, 1.50 GHz, 1.40 GHz, 1.30 GHz, 1.20 GHz
available cpufreq governors: ondemand, userspace, performance
current policy: frequency should be within 1.20 GHz and 2.00 GHz.
The governor "ondemand" may decide which speed to use
within this range.
current CPU frequency is 1.20 GHz.
You can dynamically change the governor used under the OS Control mode by modifying the value
in the /sys/devices/system/cpu/cpu*/cpufreq/scaling_governor file for each CPU.
RHEL 6.1 provides the cpufreq-set command to select the governor. For more information about
the cpufreq-util and cpufreq-set commands, refer to the RHEL 6.1 man pages.
Collaborative Power Control with RHEL 6.1
When ProLiant servers are under OS Control mode for power management, power capping may still
be imposed by the platform without the knowledge of the operating system. First introduced on Intelbased G6 ProLiant servers and included in all Gen8 ProLiant servers is the ability for the server and
the OS to collaborate on power management. HP provides the Collaborative Power Control (CPC)
mechanism which is capable of providing capping related feedback to the operating system, as well
as collaborating with the operating system in managing the power consumption of a server. This
combination provides the quick response time of HP Dynamic Power Savings and still provides correct
processor power information to the operating system.
9
CPC utilizes the Processor Clocking Control (PCC) interface which is an interface for coordinating
processor performance between the platform firmware and the operating system. The PCC interface,
jointly developed by HP and Microsoft is publicly available, allowing other platform vendors the
option of implementing it.
For more information on PCC, see:
http://www.acpica.org/download/Processor-Clocking-Control-v1p0.pdf
Platform firmware releases for Intel-based ProLiant G6 servers from August 2009 and later include
support for CPC.
When a CPC-enabled server is configured in HP Dynamic mode, the firmware does not present P-state
information to the operating system. Instead, the firmware presents the minimum and maximum
frequencies the processor supports, allowing the OS to pick any frequency within that range, rather
than restricting the processor to specific P-states. As with OS Control mode, the operating system
governor provides policy and requests through the new PCC driver (pcc-cpufreq) in RHEL 6.1 what
performance is required from the processor. The platform firmware strives to honor the requested
frequency. If the processor is capped at that time for any reason, then the platform firmware will
inform the OS that the request could not be accomplished due to capping. When capping is not
configured, the PCC driver still continues to function in lieu of the P-state driver. Example 2 shows a
sample output for CPU 0 for a Gen8 server under RHEL 6.1.In this example, notice that the driver is
pcc-cpufreq. Only the minimum and maximum frequency limits are displayed. Unlike under OS
Control, there are no preset frequency steps.
Example 2 Output for CPU 0 in HP Dynamic mode with CPC enabled
# cpufreq-info -c 0
cpufrequtils 007: cpufreq-info (C) Dominik Brodowski 2004-2009
Report errors and bugs to cpufreq@vger.kernel.org, please.
analyzing CPU 0:
driver: pcc-cpufreq
CPUs which run at the same hardware frequency: 0
CPUs which need to have their frequency coordinated by software: 0
maximum transition latency: 0.00 ms.
hardware limits: 1.20 GHz - 2.00 GHz
available cpufreq governors: ondemand, userspace, performance
current policy: frequency should be within 1.20 GHz and 2.00 GHz.
The governor "ondemand" may decide which speed to use
within this range.
current CPU frequency is 1.20 GHz.
Idle Power States (C-States) with RHEL 6.1
Processor power use at idle is a crucial factor in determining power consumption of a server when
there is no workload to execute. Typically, when a processor does not have work to perform, the
operating system places the processor in a halt state signified as C1. Newer generation processors
support deep C-states, allowing RHEL 6.1 to take advantage of these states. Although C-states can
significantly reduce power consumption, the drawback of going to a deeper C-state is the latency
associated with the time it takes for the processor to wake up from its idle state and resume executing
instructions. Information about the C-states for system processors is available in
/sys/devices/system/cpu/cpu*/cpuidle/state*.
10
NOTE: You can configure the server to not utilize the idle C-states by choosing the No C-states
setting in RBSU.
Additional RHEL 6.1 Power Management Features
RHEL 6.1 provides a comprehensive set of features for managing the power usage of ProLiant servers.
The “Green IT” features introduced in RHEL6.02 and later offer the user a range of kernel and userspace features to manage server power consumption. With the “tickless when idle” kernel feature, it is
possible to reduce the number of wakeups per seconds from 1024 to typically less than 30. For
instance, in Figure 7, notice that the “Wakeups-from-idle per second” is below 23. Additional tools
are available to monitor the system power consumption. For example, utilizing the powerTOP3 tool,
you can identify processes that are most responsible for waking a processor up from its idle state and
thereby driving up power consumption. Reference the powerTOP documentation for further reading
on what the output of powerTOP represents, and for tips and tricks on how to best tune the server for
maximum power savings.
Figure 7 displays the powerTOP output on an idle 1P ProLiant DL360e Gen8 platform with Intel(R)
Xeon(R) CPU E5-2420 processor and 2 GB system memory. The average residency in the deepest
supported C-state4 is about 49ms. This value is due to the processor being awakened about only 23
per second times from its idle state. The output listing is for a case where the IPMI service has been
stopped on the server 5.
2
3
4
5
Red Hat Enterprise Linux 6: Green Computing Features
An introduction to powerTOP: http://www.lesswatts.org/projects/powertop/powertop.php.
ACPI C3 actually corresponds to hardware C6 state which is the deepest C-state supported by the processors on that platform
Halting the IPMI driver results in a user losing the ability to remotely monitor the server. If the IPMI service is stopped in order to save power, it is
possible to resume the IPMI service with the command service ipmi restart
11
Figure 7: PowerTOP output on an idle ProLiant DL360e Gen8 with no IPMI service
NOTE: You can see the P-states (frequencies) information in the powerTOP output only when the HP
Power Regulator is configured to OS Control mode.
Summary
HP ProLiant servers are enabled for saving power both when the server is under load and when the
server is idle. The processor-based power management features when supported in the hardware are
enabled by the firmware automatically, and are used in close co-ordination between the firmware
and the RHEL 6.1 operating system. Typically, you do not have to activate these features. They are
already enabled by default. Several innovative features such as the HP Power Regulator, HP Power
Capping, HP Dynamic Power Capping, and Collaborative Power Control provide advanced power
saving and budgeting features on HP ProLiant servers.
12
For more information
For additional information, refer to the resources listed below.
Resource description
Web address
HP ProLiant Gen8
www.hp.com/go/proliantgen8
HP Linux documentation
www.hp.com/go/linux-docs, select HP Linux Server
Management Software.
Power Regulator
http://www.hp.com/servers/power-regulator
Power Capping and HP Dynamic Power
Capping
http://h20000.www2.hp.com/bc/docs/support/SupportManual/c
01549455/c01549455.pdf
Power Regulator for ProLiant servers
http://h20000.www2.hp.com/bc/docs/support/SupportManual/c
00300430/c00300430.pdf
AMD PowerNow! ™ Technology
http://www.amd.com/us/products/technologies/amd-powernowtechnology/Pages/amd-powernow-technology.aspx
Enhanced Intel SpeedStep ® Technology
and Demand Based Switching on Linux
http://softwarecommunity.intel.com/articles/eng/1611.htm
Linux cpufreq kernel documentation
http://lxr.linux.no/linux+v2.6.32/Documentation/cpu-freq/
Linux cpuidle kernel documentation
http://lxr.linux.no/linux+v2.6.32/Documentation/cpuidle/
Intelligent Platform Management
Interface (IPMI)
HP Integrated Lights-Out Management
Controller
HP Insight Power Manager (IPM)
http://www.intel.com/design/servers/ipmi/
HP System Insight Manager (SIM)
http://www.hp.com/go/sim
An introduction to powerTOP
http://www.lesswatts.org/projects/powertop/powertop.php
LessWatts
http://www.lesswatts.org/
Red Hat Enterprise Linux 6: Green
Computing Features
https://tw.redhat.com/about/news/archive/2010/5/red-hatexpands-green-computing-features-in-red-hat-enterprise-linux-6
Linux on HP ProLiant servers
http://www.hp.com/go/proliantlinux
RHEL on HP servers
http://h18004.www1.hp.com/products/servers/linux/redhat/rhel
/index.html
Processor Clocking Control (PCC)
interface
http://www.acpica.org/download/Processor-Clocking-Controlv1p0.pdf
Kernel documentation on the Linux PCC
implementation
http://lxr.linux.no/linux+v2.6.34/Documentation/cpu-freq/pcccpufreq.txt
Introduction to PCC as presented at the
Linux Foundation Collaboration Summit
(LFCS) 2010
https://events.linuxfoundation.org/slides/lfcs2010_garbee.pdf
http://www.hp.com/go/ilo
http://www.hp.com/go/ipm
Call to action
Send comments about this document to TechCom@HP.com
Share with colleagues
© Copyright 2012 Hewlett-Packard Development Company, L.P. The information contained herein is subject to
change without notice. The only warranties for HP products and services are set forth in the express warranty
statements accompanying such products and services. Nothing herein should be construed as constituting an
additional warranty. HP shall not be liable for technical or editorial errors or omissions contained herein.
Microsoft, Windows, Windows NT, Windows XP, Windows Mobile, and Windows are U.S. registered
trademarks of Microsoft Corporation.
VMware ESX, ESXi and VMotion are U.S. registered trademarks of VMware, Inc.
5900-2215, March 2012
Was this manual useful for you? yes no
Thank you for your participation!

* Your assessment is very important for improving the work of artificial intelligence, which forms the content of this project

Download PDF

advertising