Loading a table

You can improve performance on large table loads by using parallelism. In one case, we had an existing script trying to load 130 million rows into a table without using parallelism. The load was taking forever, so here is what we did to improve this load:

• Considered dropping indexes before the load

• Ensured we could use intra-partition parallelism

• Assigned memory parameters to be automatic

• Checked table space assignment

• We made sure, in this case, that we had separate table spaces for the data and index (and LOBs when present)

• Assigned buffer pools to the data, index, and temp table spaces

• Added parallelism to the load command

  nohup db2 "load from tab415.ixf of ixf
    messages tab415.msg
    replace into DTA.FLRG_TBL
    nonrecoverable
    cpu_parallelism 8
    allow no access" &
  

See also

The I/O Tuning recipe, in this chapter