II14213: USE OF DFSORT BY DB2 UTILITIES (CONTINUED FROM II14047)

APAR status

• Closed as canceled.

Error description

• THIS APAR IS A CONTINUATION OF II14047. PLEASE ENSURE THAT YOU
  HAVE REVIEWED ALL OF THE INFORMATION IN APAR II14047 REGARDING
  THE USE OF DFSORT BY DB2 UTILITIES.
  .
  3. TAILORING DFSORT INSTALLATION DEFAULTS TO YOUR ENVIRONMENT
  .
  DFSORT supports various options for four invocation environments
  (JCL, INV, TSO and TSOINV) and four time-of-day environments
  (TD1-4).  With DFSORT as your primary sort product, you can
  customize the DFSORT options for these environments as
  appropriate.  But when you have another primary sort product and
  are just using DFSORT for DB2 Utilities, the only environment
  you can change is the INV environment.  DFSORT is shipped with
  default options for this environment that generally provide
  efficient operation.  However, you may want to change some of
  the INV options to better match your own environment for running
  DB2 Utilities.  In particular:
   - you may want to increase the DSA value above the shipped
     default of 64 (megabytes) for z/OS DFSORT V1R5 or 32
     (megabytes) for DFSORT R14, e.g., DSA=128
   - you may want to increase the number of dynamically allocated
     work data sets above the shipped default of 4, e.g.,
     DYNALOC=(SYSDA,8).  A "best practice" here is to set this
     number to a value that satisfies 80-90% of your sorts, and
     then use the SORTNUM parameter on those utilities that
     require a different value.
   - you may want to change the shipped default of DYNAUTO=YES to
     DYNAUTO=IGNWKDD to allow DFSORT to automatically use
     dynamically allocated work data sets instead of only the JCL
     work data sets.
   - you may want to change the shipped default of SMF=NO to
     SMF=SHORT or SMF=FULL to record DFSORT SMF type 16 records.
  .
  See "z/OS DFSORT Installation and Customization" and "z/OS
  DFSORT Tuning Guide" for more information.  You can access these
  and the other DFSORT books on-line using the "Publications" link
  on the DFSORT website at:
  http://www.ibm.com/storage/dfsort
  .
  To modify your DFSORT installation options, write a usermod to
  modify ICEMAC. The following instructions illustrate the process
  for changing the DSA value to 128 (megabytes) for use by DB2
  Utilities.
  .
  The sample library SICESAMP contains the following members:
   - ICEOPREC does the RECEIVE
   - ICEOPAPP does the APPLY
  .
  Update the ICEMAC INV statement in the ICEAM2 module within the
  ICEOPREC sample job
  ++ SRCUPD (ICEAM2) DISTMOD (AICELIB) DISTLIB (AICESRCE)
   /* The ICEAM2 member in the AICESRCE library contains the
      default lines for creating the ICEAM2 module. A model ICEMAC
      INV statement for customizing the ICEAM2 module is shown
      below. Replace it with an ICEMAC INV statement that
      specifies the parameters you want to use for ICEAM2.
      You MUST use the correct sequence numbers in columns 73-80
      to replace the appropriate lines in the ICEAM2 member.
   */ .
  ./ CHANGE NAME=ICEAM2                                  73
     ICEMAC      INV,DSA=128                              00550000
  .
  You can remove the ICEAM1, ICEAM3, ICEAM4 updates in the
  ICEOPREC sample job. Ensure the FMID is correct (HSM1G00 for
  DFSORT R14, or HSM1H00 for DFSORT V1R5). Also ensure you use the
  same SMP target zone and options you use for installing z/OS
  products.  Then do the RECEIVE using the modified ICEOPREC
  sample job and the APPLY using the ICEOPAPP sample job. When you
  install a new release of z/OS, this usermod has to be RECEIVEd
  and APPLYed to the new release.
  .
  4. VERIFYING ACS ROUTINES DO NOT ALLOCATE DFSORT WORK DATA SETS
  TO VIO
  .
  The use of Virtual I/O (VIO) for DFSORT work data sets is NOT
  recommended. VIO=NO is DFSORT's shipped installation default and
  recommended setting for the VIO parameter. If your site's ACS
  routines allocate DFSORT work data sets to VIO, it is
  recommended that you change your ACS routines to avoid using VIO
  for SORTWK*, STATWK*, DATAWK*, DAnnWK*, STnnWK* and SWnnWK* data
  definitions.  This is because DFSORT already performs storage
  hierarchy management (the efficient balancing of processor
  memory and disk storage), and mixing this with VIO which uses
  processor memory results in suboptimal efficiency.
  
  .
  DFSORT exploits processor storage through the use of either
  Hiperspaces, Dataspaces or Memory Objects.  Memory Object
  sorting will only be used if MEMLIMIT is non-zero.  If MEMLIMIT
  is not set in your JCL then it defaults to the MEMLIMIT value
  specified in the SMFPRMxx member of SYS1.PARMLIB.  If MEMLIMIT
  is not specified in SMFPRMxx then that default is zero.  More
  information on the relationship between MEMLIMIT, REGION, IEFUSI
  and SMFPRMxx can be found in the z/OS MVS Programming:  Extended
  Addressability Guide.
  .
  DFSORT uses a primary allocation of 0 tracks for the initial
  allocation of work data sets.  This can impact the decisions
  made by DFSMS Automatic Class Selection (ACS) routines if the
  &size variable is being used to assign a unit (such as VIO) or
  select a storage class and group based on data set size. It is
  recommended that the &maxsize variable be used as a basis for
  decisions related to data set size. Since DFSORT work data set
  allocations contain a secondary space quantity, the &maxsize
  variable will allow for a more accurate decision on the
  assignment of unit, storage class, storage group, etc.
  .
  While DFSORT performs the initial work data set allocations with
  a primary space quantity of 0, it will attempt to extend them
  once the accurate amount of required space is determined.
  Should a work data set reside on a volume which does not have
  sufficient space available, DFSORT invokes recovery routines
  that may result in the data set being deleted and re-allocated.
  .
  5. EXCLUDING DFSORT WORK DATA SETS FROM OEM PRODUCT'S USE
  .
  Exclude SORTWK*, STATWK*, DATAWK*, DAnnWK*, STnnWK*,
  and SWnnWK* datasets from any products that
  reduce space allocations.
  .
  While running concurrent utilities, if the percent free of your
  work pool is low, you can either reduce the number of
  concurrently running utilities or you can increase the number of
  volumes in your work pool.  If you increase the number of
  volumes, you can then monitor utilization and bring that number
  back down if it's consistently underutilized.   As a general
  rule, your work pool should average over 20% free to allow for
  optimal allocation of DFSORT work data sets and reduce the risk
  of failures due to lack of work space (ICE083A, ICE046A, etc.)
  .
  6. REGION SIZE (MAIN MEMORY) OF THE DB2 BATCH UTILITY JOB
  .
  DFSORT may abend or process inefficiently for large sorts
  when there is not enough main memory available, which can
  occur especially when the Utility runs many sorts in parallel.
  For large sorts when main memory is constrained, DFSORT will
  compensate with intermediate merges that increases the need for
  sort work data sets.  One method to avoid this is to ensure that
  there is sufficient main memory available for large sorts.  The
  following information gives a rough estimate on the amount of
  memory that DFSORT would need to run efficiently:
  .
  Data size       Estimated main memory use per DFSORT task
     1 GB                         10 MB
    10 GB                         30 MB
   100 GB                         70 MB
  where data size is the number of bytes to be sorted, e.g.,
  record count * record length.
  .
  Increasing the region size of the batch Utility job to
  accomodate the main memory needed for the parallel DFSORT tasks
  can allow the sorts to succeed.
  .
  7.LIMITING THE NUMBER OF PARALLEL DFSORT TASKS FOR DB2 UTILITIES
  .
  The number of parallel DFSORT tasks for Utilities may be limited
  by allocating the UTPRINxx data sets to something other than
  SYSOUT *.  E.g., if UTPRIN01, UTPRIN02, and UTPRIN03 are
  allocated to real data sets then at most 3 parallel sort tasks
  would be initiated by the Utility.
  .
  Service
  -------
  Even if you do not have a license for DFSORT, IBM will provide
  central service for DFSORT problems related to the use of DFSORT
  by DB2 Utilities. IBM already provides all DFSORT service (PTFs)
  to all z/OS customers. If a DFSORT defect is encountered, DFSORT
  service and any prerequisite PTFs will be required.  You must be
  aware that you now "care" about DFSORT service and take
  appropriate actions. For example, IBM recommends that you
  specify DFSORT in the group of FMIDs that you want services. If
  you use Enhanced HOLDDATA, DFSORT HIPERs will show up in the
  REPORT ERRSYSMODS listing.  If you've tailored notifications
  (like ASAP in ServiceLink) at the FMID level, you should add
  notifications for DFSORT HIPERs.
  .
  If you have a any problem with DFSORT (e.g., DFSORT messages
  ICE039A, or ICE046A, or you receive an abend), you should open a
  PMR against DB2. If necessary, DB2 support will transfer the
  problem to DFSORT. If the problem can be reproduced, add
  SORTDIAG and DFSPARM DD statements to the failing job and rerun
  it:
  //SORTDIAG  DD   DUMMY
  //DFSPARM   DD   *
     DEBUG ABEND
  /*
  .
  This will allow DFSORT to display additional diagnostic (ICE8xxx
  and ICE9xxx) messages, and ABEND if a DFSORT error message is
  issued.  Send the complete JOBLOG and dump (if an error occurs)
  to the appropriate IBM service representative.
  .
  Recommended DFSORT service:
  ++APAR  PK25047 - ICE046A can result when DFSORT uses multiple
  Hiperspaces and more than 2G of hiperspace is required to
  successfully complete the sort.
  .
  Additionally, there are a number of APARs available for both
  DFSORT and DB2 that affect the way they interact. You should
  ensure that you are current on both DFSORT and DB2 maintenance.
  

Local fix

Problem summary

Problem conclusion

Temporary fix

Comments

• This info apar is being closed so it is searchable
  by customers.
  

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