VM65686: FCX100 SHOWS EXCESSIVE VALUES FOR ONE PROCESSOR

Obtain the fix for this APAR.

APAR status

• Closed as program error.

Error description

• Perfkit FCX100 CPU Load and Transactions Report has excessive
  CPU values due to errors in Monitor D0R2 field SYTPRP_PFXTOTWT.
  The values reported in Domain 0, Record 2 SYTPRP_PFXTOTWT
  occasionally regress, reporting less cumulative wait time than
  the prior record.
  

Local fix

Problem summary

• ****************************************************************
  * USERS AFFECTED: Customers using Monitor or Perfkit           *
  ****************************************************************
  * PROBLEM DESCRIPTION:                                         *
  ****************************************************************
  * RECOMMENDATION: APPLY PTF                                    *
  ****************************************************************
  Perfkit FCX100 CPU Load and Transactions Screen reports CPU
  values that are excessive when values reported in Monitor
  Domain 0, Record 2 SYTPRP_PFXTOTWT regress, reporting less
  cumulative wait time than was reported in the prior record.
  
  This may occur when a processor appears to be in wait state
  when Monitor records the value SYTPRP_PFXTOTWT, but exits wait
  before Monitor completes its calculations.
  
  APAR VM66105 changed the algorithm for SYTPRP_PFXTOTWT value
  reporting to improve accuracy. The accuracy problem was due to
  the fact that the wait state timer is only updated when a
  processor exits wait state. If the processor was in wait state
  for an entire Monitor interval, then SYTPRP_PFXTOTWT would
  not increase for that interval, but would jump ahead by a
  large mount when the processor eventually exited wait state.
  APAR VM66105 corrected that problem by having Monitor adjust
  the reported value by calculating the elapsed time since the
  processor entered wait state.
  
  The problem occurs when Monitor sees the processor as in wait
  state just as that processor is exiting wait state. Monitor
  calculates a delta to add to SYTPRP_PFXTOTWT which is larger
  than the delta that the waking processor wait timer actually
  changed. In the next interval Monitor sees the processor is
  not in wait state and reports an unmodified wait timer value
  in SYTPRP_PFXTOTWT which indicates less cumulative wait time
  than it reported in the previous interval.
  
  SYTPRP_PFXTOTWT is reported in CPU Timer format so that the
  value decreases as wait time increases. In the case of the
  reported problem the SYTPRP_PFXTOTWT value increased in the
  next interval which corresponds to less cumulative wait time.
  
  In customer supplied data and experiments to recreate the
  problem, the amount of regression was in the range of a few
  milliseconds. Although this may seem insignificant, the
  calculation of the difference between the values from two
  intervals ends up as a negative value. Perfkit should never
  get a negative value so it is unprepared for that case and the
  value is treated as a very large unsigned number instead.
  

Problem conclusion

• The smallest possible Monitor interval is 6 seconds. As noted
  earlier, the amount of regression in the wait time value is
  very small.  If the error was even as large as 10 milliseconds
  then the error would be roughly 0.16%.
  
  To prevent the SYTPRP_PFXTOTWT value from regressing, Monitor
  will remember the value that was reported for the prior
  interval to determine whether an adjustment is needed when
  it next reports SYTPRP_PFXTOTWT.
  
  Code changes were made to:
  HCPPLSBK COPY - added PLSTOTWT for Monitor to remember what was
                  reported in SYTPRP_PFXTOTWT for each processor.
                - added PLSTOTWTLO to use to track the largest
                  amount of regression in SYTPRP_PFXTOTWT values.
  HCPMNY PLX    - uses PLSTOTWT and PLSTOTWTLO this way:
                  a. PLSTOTWT = 0 indicates it has not been used
                     and needs to be initialized so set
                     PLSTOTWT to newly calculated SYTPRP_PFXTOTWT.
                  b. PLSTOTWT < newly calculated SYTPRP_PFXTOTWT.
                     Monitor reported more total wait time last
                     interval. Set SYTPRP_PFXTOTWT = PLSTOTWT.
                     to report the same value in this interval.
                     Also calculate the amount of regression
                     avoided. If that value is larger than the
                     current PLSTOTWTLO value then set PLSTOTWTLO
                     to the newly calculated difference.
                  c. PLSTOTWT > newly calculated SYTPRP_PFXTOTWT
                     then the processor has been in wait state
                     longer than was last reported. Therefore set
                     PLSTOTWT to newly calculated SYTPRP_PFXTOTWT
                     to remember the new value.
  

Temporary fix

• FOR RELEASE VM/ESA CP/ESA R640 :
  PREREQ: VM66105
  CO-REQ: NONE
  IF-REQ: NONE
  FOR RELEASE VM/ESA CP/ESA R710 :
  PREREQ: NONE
  CO-REQ: NONE
  IF-REQ: NONE
  

Comments

APAR Information

• APAR is sysrouted FROM one or more of the following:

• APAR is sysrouted TO one or more of the following:

  UM35191 UM35270

Modules/Macros

• HCPMNY   HCPPLSBK
  

Fix information

• Fixed component name

  VM CP

• Fixed component ID

  568411202

Applicable component levels

• R640 PSY UM35191

     UP18/10/10 P 1802

• R710 PSY UM35270

     UP18/10/10 P 1901

Fix is available

• Select the PTF appropriate for your component level. You will be required to sign in. Distribution on physical media is not available in all countries.