AIX LPAR CPU CAPACITY PLANNING

If AIX LPAR is using Shared CPU ( not dedicated CPU ), you can examine this guidance.

			Feel free to communicate by bulent.yucesoy@gmail.com




It is advised that you enable "Allow performance information collection"  checkbox at HMC.
So that, you will be able to retrieve APP (Available Physical Processor) data at "LPARSTAT 1" command telling CPU POOL USAGE.

"LPARSTAT 1" command has a field called "LBUSY".
It means "LOGICAL BUSY"
At the end of all PowerVM CPU terminology ( EC,VC, PC, SMT etc...), OS sees LOGICAL CPUs.
LBUSY gives average cumulative CPU usage of all your OS LOGICAL CPUs.



  What you will examine is "LPARSTAT 1" command output.
 
   - APP must not be near 0 at first. If it is like that, you must examine your cpu pool and LPARs inside that pool.
     We care about single LPAR Capacity planning here, so we assume that there is no bottleneck due to unavailable cpu at cpu pool.
     Also if you make each single LPAR analysis correctly, you can group systems and make cpu pool designs also better.

   - You must record your "physical cpu usage" + "LBUSY" values historically all the time.
   
     * "Physical cpu" usages when "LBUSY was near 50" ( it means % 50 ) are your average physical cpu usages and EC value can be set due to this.

     * "Physical cpu" usages when "LBUSY was between 90 and 100" may be considered as your peak physical cpu usages and plans can be done due to this.



  CPU bootleneck at LPAR is something different than capacity planning but if short notes needed for CPU bottleneck detection,

    - If there is high cpu utilization at LPAR like %90 for example, LBUSY will be 90. LBUSY is important.

    - vmstat "r" data must not be so much bigger than total logical cpu count. Eg: lcpu = 16, r=50. It is problematic

    - High load averages may also point cpu bottleneck. My personal observation tells that loads may not be around 40, must be less.