Question - Can we reduce our IFL count?

We are looking at the possibility of reducing the number of IFLs. What can I look at to evaluate this possibility?

Note: This is a guideline of 'where to look' and 'what to look for' to assist in research. Asking for assistance from Velocity Software before making any decisions is recommended.

• Look at Totals by Processor Type at the bottom of each daily report. Average the numbers for at least a month, over busy periods and peak times/days.
• If 80% or above, keeping all the IFL's is a good idea.
• If it is below 80%, IFLs can possibly be removed without issue.
• Look at Totals by Processor Type at the bottom of each 15 minute interval. Determine the high water marks
• If any 15 minute interval is over 80%, can workloads be changed to accomodate these spikes? Would RESPOOLS be useful?
• If the 15 minute intervals for a month are all below 80%, IFLs can possibly be removed without issue.

Totals by Processor type:
<---------CPU-------> <-Shared Processor busy->
Type Count Ded shared  Total  Logical Ovhd Mgmt
---- ----- --- ------ ------ -------- ---- ----
CP       2   0      2   97.4     97.0  0.2  0.2
IFL      4   0      4  132.2    128.6  1.7  1.9 <-- For this day, this system was at approximately 33%
ZIIP     1   0      1   24.2     24.2  0.0  0.0

Totals by Processor type:
<---------CPU-------> <-Shared Processor busy->
Type Count Ded shared  Total  Logical Ovhd Mgmt
---- ----- --- ------ ------ -------- ---- ----
CP       2   0      2  160.6    157.9  1.0  1.7
IFL     15   0     15 1215.4   1209.9  3.2  2.3 <-- For this day, this system was at approximately 81%
ICF      1   1      0    0.0        0    0  0.0
ZIIP     1   0      1    5.8      5.4  0.1  0.3

• Look at the Summary of the processors at the bottom of each daily report. Average the numbers for at least a month, over busy periods and peak times/days.
• If 80% or above, keeping all the IFL's is a good idea.
• If it is below 80%, IFLs can possibly be removed without issue.
• Look at the CPU percentages for each 15 minute interval. Determine the high water marks
• If any 15 minute interval is over 80%, can workloads be changed to accomodate these spikes? Would RESPOOLS be useful?
• If the 15 minute intervals for a month are all below 80%, IFLs can possibly be removed without issue.
• If there are CPUs that are parked often, this could also indicate that fewer CPUs are needed.

        <----Load---->           CPU (percentages) .  <-Vertical-->
         <-Users-> Tran     CPU   Total  Emul      .  Entitle- Park
Time     Actv In Q /sec CPU Type   util  time      .  ment  seconds
****************Summary********************************************
Average:   77 41.9 13.4  0  IFL    18.8  17.7      .  0.37        0 <-- 4 of 6 IFLs evenly used, relatively low utilization
                         1  IFL    21.2  20.1      .  0.37        0
                         2  IFL    19.7  18.7      .  0.37        0
                         3  IFL    19.8  18.8      .  0.37        0
                         4  IFL    17.5  16.5      .  0.35     4142 <-- 2 IFLs that are being parked
                         5  IFL    17.6  16.6      .  0.35     4142
                                  ----- -----      .  ----     ----
System:                           114.7 108.5      .  0.36     8284 <-- For this day, this system was at apprixmately 20%

        <----Load---->           CPU (percentages) .  <-Vertical-->
         <-Users-> Tran     CPU   Total  Emul      .  Entitle- Park
Time     Actv In Q /sec CPU Type   util  time      .  ment  seconds
Average:   31 81.5  0.5  0  IFL    65.0  60.0      .  1.00        0 <-- 52 of 52 IFLs evenly used, healthy utilization
                         1  IFL    59.2  54.7      .  1.00        0
                         2  IFL    68.0  64.1      .  1.00        0
                         3  IFL    63.3  59.8      .  1.00        0
                        - - - - - - - - - - -
                        48  IFL    81.0  77.2      .  1.00        0
                        49  IFL    79.0  75.4      .  1.00        0
                        50  IFL    65.4  60.0      .  1.00        0
                        51  IFL    62.9  57.9      .  1.00        0 <-- No parking
                                  ----- -----
System:                            4032  3843      .  1.00        0 <-- For this day, the system was at approximately 78%

• Scan and trend the summary of the amount of users in CPU wait. Any number over 20 shows that there is an issue with users getting the CPU they need.
• If this happens sporadically, it could be workload or a user issue. (ESAUSP2 can be used to research users using execessive CPU)
• Again, if numbers over 20 are present, what needs to be changed to rectify this before considering removing IFLs.

                     <-(Wait states)->
UserID   <-Samples->
/Class   Total  In Q Run Sim CPU
Hi-Freq: 8732K 97825 1.4 0.8 0.7  <-- For this day, there was very little CPU wait - but need to look by interval as well

Conclusions from the examples above: