How to speed up your POWER server with no cost

Hardware, HMC, POWER 2 Minutes

In my previous post I’ve mentioned about the new HMC feature which can adjust CPU frequency. In this article I want to describe how to speed up the POWER system box, get more performance from the system. It is very easy to overclock the machine.

How?

For POWER6, POWER7 MMB, MMC you need IBM Systems Director with Active Energy Manager plugin, but for MMD and POWER8 models you can do it directly from the ASM (Advanced System Management) or if you run the latest HMC code 8.8.2. there are two new commands chpwrmgmt and lspwrmgmt .

What is all about? IBM gave us an opportunity to modify a CPU frequency. Either to reduce consumed energy or overclock the CPU and boost the machine speed.

How to enable it in the ASM?

Open the ASM window from the HMC console. Expand the System Configuration > Power Management > Power Mode Setup.

Power Mode Setup
Power Mode Setup
In this screen you have four options.

  • Disable Power Saver mode – this is default. No changes in the CPU frequency, resource will operate at 100% of nominal processor frequencies all time.
  • Enable Static Power Saver mode – activates the Power Saver mode on IBM Power servers, fixing the processor frequency and voltage at a predetermined low-power mode. In another words, processor speed will be reduced in order to save energy if there is no demand for high CPU.
  • Enable Dynamic Power Saver (favor power) mode – guarantees power savings by limiting the maximum frequency of the system under peak utilization under high utilization.
  • Enable Dynamic Power Saver (favor performance) mode – allows a higher frequency range under
    high utilization.

How to enable via CLI in the HMC?

To display current configuration type in the CLI

lspwrmgmt -m server_name -r sys 

To change power management setting

chpwrmgmt -m server_name -r sys -o enable

or

chpwrmgmt -m server_name -r sys -o enable -t dynamic_favor_perf

but apparently the last command doesn’t work for me. It finishes with the error “HSCL02F2 The managed system does not support setting dynamic power saver modes from the management console”. I don’t why, well basically I don’t care because I can enable the same feature from the ASM.

Note that on multi-node systems such as the Power 770 or 780, each node’s processor frequency is
controlled independently. Across an entire system, processor core frequency will be maintained within
a 10% envelope.

So if you want to get max performance from your box, and you don’t care much about the electricity select Enable Dynamic Power Saver (favor performance) mode. You may achieve about +20% boost from the nominal speed.

What is the risk?

Well, I’m not a hardware engineer, but I believe it’s very safe. Changing Power Modes is an online operation, so you can give a try. But remember, if you have a system in maximum configuration and you will run overclocking, the power consumption will be higher than described in the hardware specification. So double check with your facility stuff that power cords are able to deliver that much energy.

Great thing about the overclock is that you get an extra performance (up to 20%) from the same system. You don’t have to buy anything (if you run MMD or Power8, otherwise you need IBM Systems Director and Active Energy plugin), you also don’t brake any licenses, and you get more performance. That’s why you probably will not hear about this feature from your sales-represantive;)

If you want to read more about it, check this IBM document.

 

Published by Bart

Published

15 thoughts on “How to speed up your POWER server with no cost”

  1. Bart, in order for the Dynamic Power Mode feature to work from the HMC, two things are required. You need the V8R8.2.0 level of HMC as you stated. On the managed system side, you also need the latest service pack for the firmware code stream as it requires an updated service processor to recognize the new HMC commands (even though as you found, you can still control the dynamic power mode using ASMI on the service processor). So for the FW780 code stream, the system requires service pack 780.40 which will GA 12/16/2015. Other FW levels have already GA’ed. These are FW760.50, FW770.50, FW810.20, FW820.00, and FW 830.00. The fix is also in FW840.00 which is set to GA in December 2015.

    Reply

    1. Dieter,
      AIX and VIOS (under root shell) lsattr -E -l proc0 | grep “Processor Speed”
      for IBM i I’ve never found a way how to show the clock speed.

      However, if you enable the “Turbo Mode” , you will not see clock running with higher freq right away. The CPU is overclocked only if there is demand for it. Otherwise it runs with its nominal speed.

      Regarding the power consumption – I don’t think there is a way to see it by some command on a server. The only way is to use an external device to measure it.

      Reply

      1. Bart, one technique for measuring power consumption on P8 systems is to use ipmitool to connect to the service processor and then run DCMI commands to get the current power consumption. Here is an example where tul259fp is the FSP2 service processor:

        ipmitool -I lanplus -H tul259fp -P PASSW0RD dcmi power reading

        Instantaneous power reading: 325 Watts
        Minimum during sampling period: 325 Watts
        Maximum during sampling period: 325 Watts
        Average power reading over sample period: 325 Watts
        IPMI timestamp: Sat Feb 20 00:16:09 2016
        Sampling period: 00030000 Milliseconds
        Power reading state is: activated

      4. What you describe with “by demand for it” is no longer the case. What you describe was the “Favor Performance” setting in ASMI Power Management. I don’t remember the specific POWER8 firmware, 820 or 830 but it introduced an additional option that lets you set the frequency to the max setting where it runs all of the time. Here is a link for a L box. It is the same for all non-LC servers except the E850 which doesn’t permit any increase due to thermal constraints. http://www.ibm.com/support/knowledgecenter/POWER8/p8hby/areaa_pmms.htm

  3. It seems, that on new POWER9 servers it will be already set:
    The Power S924 server by default has its Power Management mode set to Max Performance. This mode can dynamically optimize the processor frequency at any given time based on CPU utilization and operating environmental conditions. For a description of this feature and other power management options available for this server, see the IBM EnergyScale for POWER9 Processor-Based Systems website.

    Reply

  5. Bart
    I have never used the ASM before what is the default user id to log in.
    Thanks

    Reply

Leave a reply to Dieter Cancel reply