II12033: COUPLING FACILITY PERFORMANCE PROBLEMS CAUSED BY CF CONFIGURATION OPTIONS

APAR status

• INTRAN

Error description

• This INFO APAR documents Coupling Facility performance
  problems that can be caused by using various CF configuration
  options.  Although some such options may be helpful for certain
  situations, they are inappropriate for production environments.
    Coupling Facilities can be defined to use shared or dedicated
  engines (either CP, ICFs or combinations of the two).
  Dedicated engines give the best performance.  If multiple CF
  partiitons share engines, each partition receives only a portion
  of the CF CPU resource. The CF microcode (CFCC) runs in a
  polling loop, looking for work.  So even a test CF partition
  with little activity looks just as busy as a production CF.  For
  example, a production CF and a test CF sharing 1 ICF, would
  each only have use of 1/2  ICF.  This can be seen in the RMF
  report as DEFINED 1, EFFECTIVE 0.5.    Requests to a CF
  that is sharing a CP may have to wait if the other CF is using
  it.  This has the net effect of elongating service times, some-
  times by a factor or 5 to 10 times.
    In the case where multiple CFs are sharing CPs or ICFs, the
  installation can provide more CPU resource to the
  production CF by using the Dynamic Dispatch option.  (This
  command is issued from the CF Console).  If you turn
  DYNDISP ON for the non-production CF, the polling loop will
  periodically pause.  The number and length of the pauses is
  determined by the actvitiy in the non-prodution CF.  While this
  gives more CPU resource to the production CF, it elongates
  the service time in the non-production CF,  in most cases by
  a factor of 100 or more.  Clearly, production CFs should have
  DYNDISP OFF.
    To display the state of CF Dynamic Dispatch:
  Select the LPAR and go to Operating System Messages. Type
  in "D DYNDISP" on the command line and click on the
  "SEND COMMAND button on the bottom of the screen. The
  response will tell you if the feature is enabled or
  disabled.
    If you enable Dynamic Dispatch, you may also observe negative
  LPAR managment time. This can occur when a low-weighted MVS
  image issues a send message to a CF with DYNDISP on.  The time
  to complete the command may exceed the timeslice for the low-
  weighted MVS image.  Even in the case where the MVS is not con-
  nected to the CF, messages to the CF occur because polling and
  monitoring are done down every CHPID that is configured online
  to the CF even if that CF is not in the active policy.  If you
  share engines so that service times are elongated, you may want
  to configure the CHPID's offline until such a time as they are
  needed.   This would eliminate the polling activity and the
  negative LPAR management time.
  

Local fix

Problem summary

Problem conclusion

Temporary fix

Comments

APAR Information

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

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

Fix information

Applicable component levels