HMC

 View Only
Expand all | Collapse all

HCM PCM understanding server_procesor and server_physicalProcessorPool

  • 1.  HCM PCM understanding server_procesor and server_physicalProcessorPool

    Posted Wed May 08, 2024 12:40 PM
    Edited by Piper Wilson Thu May 16, 2024 11:20 AM

    Hello,

    I have concern regarding 2 values from HMC. I am working on presenting the data from HMC using REST API (Nextract plus).

    I cannot understand difference between 2 values

    1) server_processor - utilizedProcUnits

    2) server_physicalProcessorPool utilizedProcUnits

    Bellow we are looking on comparison of those 2 value on P9 Frame (40CPU). This yellow one is representing what lpar2rrd represents as CPU Pool, however I cannot understand what is a green one - server_processor - utilizedProcUnits and how can I use it for monitoring.

    Looking for example to understand how it works, thank you! :)


    ------------------------------
    Marcin Michalowski
    ------------------------------


  • 2.  RE: HCM PCM understanding server_procesor and server_physicalProcessorPool

    Posted Fri May 10, 2024 07:32 PM
    Edited by Piper Wilson Thu May 16, 2024 11:20 AM

    Hi Marcin,

    This is something that doesn't give me peace of mind either:)

    I have noticed that the utilization of "utilizedProcUnits" taken from "server_processor" matches the sum of the "utilizedProcUnits" utilization of LPARs on the machine (left side of the screenshot), and the utilization of "utilizedProcUnits" taken from "server_physicalProcessorPool" corresponds to the sum of the LPAR utilization expressed as ("utilizedUnCappedProcUnits" + "utilizedCappedProcUnits").
    I think the similarity is clearly visible in the screenshot.

    It is worth noting that documentation says: (link)
    utilizedProcUnits = utilizedUnCappedProcUnits + utilizedCappedProcUnits - idleProcUnits

    If you are use HMC older than V10R2M1040 then the formula is different:(link to old documentation)
    utilizedProcUnits = utilizedUnCappedProcUnits + utilizedCappedProcUnits

    Therefore, my hypothesis is that in the case of "server_processor" the CPU Idle, ("idleProcUnits") is subtracted, but I would like to see an explanation from someone smarter than me :)


    Queries which I used:

    • Left side:
      SELECT "utilizedProcUnits" FROM "lpar_processor" WHERE ("servername" =~ /^$Servername$/) AND $timeFilter GROUP BY "lparname"
      SELECT "utilizedProcUnits" FROM "server_processor" WHERE ("servername" =~ /^$Servername$/) AND $timeFilter 

    • Right side:
      SELECT "utilizedCappedProcUnits" + "utilizedUncappedProcUnits" FROM "lpar_processor" WHERE ("servername" =~ /^$Servername$/) AND $timeFilter GROUP BY "lparname"
      SELECT "utilizedProcUnits" FROM "server_physicalProcessorPool" WHERE ("servername" =~ /^$Servername$/) AND $timeFilter 




    Best regards,
    Michal Wiktorek



    ------------------------------
    https://www.linkedin.com/in/michal-wiktorek-83b2b47b
    ------------------------------

    Original Message


  • 3.  RE: HCM PCM understanding server_procesor and server_physicalProcessorPool

    Posted Wed May 22, 2024 12:35 PM

    Hi Marcin,

    it look like something what will be included in next LPAR2RRD release as well (June 2024).

    It is CPU usage from OS point of view: CPU physical allocation (what LPAR2RRD presents now) - CPU idle (OS idle cycles)

    It has been already implemented in our new tool called XorMon Next Generation, check some detailed explanation with examples https://xormon.com/IBM-Power-CPU-monitoring.php



    ------------------------------
    Pavel Hampl
    ------------------------------

    Original Message