II14026: PERFORMANCE DEGRADATION AFTER CONVERTING LLA MANAGED PDS TO LLA MANAGED PDSE

APAR status

• Closed as canceled.

Error description

• Note: The text of this APAR was updated on 04/09/2012 and
  again on 02/25/2015 to clarify the effect that page fault
  loading has upon LLA and VLF caching, to provide a better
  explanation for how the Binder FETCHOPT parameter affects
  this processing, and to remove incorrect and/or extraneous
  information.
  This update supersedes and corrects the related discussion
  in Redbook SG24-6106, the Partitioned Data Set Extended
  Usage Guide, which contained information based upon the
  original text in this APAR.
  ******************************************************
  Some customers have reported higher I/O rates or performance
  issues after migrating modules from a PDS to a PDSE, because
  the modules were no longer being cached in VLF by LLA even
  though the PDSE library was LLA managed. It was determined
  that the reason that the PDSE program objects (modules) were
  not cached was that the use of page fault loading caused LLA
  to decide that they were not good candidates for the VLF cache.
  -
  Page fault loading is a method chosen by the Loader for large
  modules, where only part of the program is loaded into storage
  initially and the rest is brought in as needed via paging
  operations.  Since only part of the program is loaded into
  storage initially, LLA's staging algorithm may decide that
  the program would not benefit from being cached in VLF.
  Avoiding page fault loading will help LLA to make the best
  decision about whether a module should be cached in VLF.
  -
  The Loader selects which loading method to use based upon the
  the characteristics of the program object to be loaded.  Page
  fault mode may be chosen by the Loader for large program
  objects when *none* of the following conditions exist:
  . the program object is to be fetched from the LNKLST
  . the program object is to be fetched from an HFS
  . the program object is to be loaded to Global (common storage)
  . the program object is to be loaded to a user-supplied address
  . the program object is less than 96k in size
  . the program object was bound with FETCHOPT=(NOPACK,PRIME)
    or FETCHOPT=(PACK,PRIME)
  -
  Binder option FETCHOPT=(NOPACK,PRIME) causes the Loader to
  bring the entire module into storage immediately, and is a
  recommended way of ensuring that the program is not loaded in
  page fault mode if none of the other conditions listed above
  are fulfilled.  This option does not guarantee that a program
  will be cached in VLF.  It only helps LLA to make the best
  determination as to whether caching the program object in VLF
  is appropriate.  Binder option FETCHOPT=(PACK,PRIME) also will
  ensure that page fault loading is not used, but NOPACK should
  provide better performance when PACK is not otherwise required.
  -
  The CSVLLIX1 and CSVLLIX2 exits can be used to tell LLA to
  cache individual modules or modules from specific libraries.
  -
  Note that when a program object resides in a library that
  is not LLA managed, the default Binder option of
  FETCHOPT=(NOPACK,NOPRIME) may provide better performance.
  

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