Upgrade Guide for DB2 UDB for z/OS > Tuning the Production Upgrade Scripts >

Optimizing Unload and Load Job Performance

Upgrades: All upgrades.

Environments: Production test environment only. Does not apply to production environment.

This topic describes ways in which you can improve the performance of the unload and load jobs for the production upgrade.

• Optimize the unload and load jobs to reach maximum parallelism:
  • Run as many of the unload / load jobs in parallel as the DB2 subsystem can support.
  • Change the generated REXX exec job submission order to submit the longest running unload/load jobs first.

    If all the unload jobs are run in parallel, the shortest amount of time this process can take is the length of time it takes for the longest unload job to complete.

  • For partitioned tables, split the unload files so that data is unloaded and loaded in parallel for each partition. Add the WHERE clause to the unload SQL to control the data that is unloaded.
  • Overlap load and unload jobs.

    Once an unload job for a table has completed, the load job for that table can be started (assuming you have a one-table per table space database schema structure). This means that load jobs can be running at the same time as unload jobs.

• Add the ORDER BY clauses to the unload SQL to load data in clustering sequence (you must manually add ORDER BY clauses to the unload SQL).

  The *.pretedit.jcl (pretkeys) job builds ORDER BY clauses for individual tables into the data set *.syskeys.orderby.

• Use third-party utilities to accelerate the unload/load process.

  NOTE:  You are required to contact Siebel Expert Services before using third-party utilities.

  You can use the DB2 Cross Loader (an option of the IBM Load utility) to load data directly from the source to the target database, thereby eliminating the unload step.

• Populate new columns as part of the unload SQL.

  NOTE:  You are required to contact Siebel Expert Services before doing this.

• Add any large tables to *.TABLIST so that the unload and load processes use the large proc, SIEBEL.PROC(SV7LD10L), which allocates more memory, instead of using the standard proc, SIEBEL.PROC(SV7LD10S).