Hi All;
I have a 2.5mm row table (very wide columns) on an UDB v8.2.3 (DPF -
6x10) AIX box. We were doing some performance testing to determine the
pros and cons of delete/insert vs load from cursor operations in our
ETL architecture (data warehouse extract/transform/load).
We expected load from cursor to be substantially faster, but found that
the delete/insert took 2 minutes and the load from cursor took 5
minutes.
The delete portion of the delete/insert test deletes all rows and the
insert performs the same SELECT as the load from cursor which does a
load/replace.
Can anyone help me understand why this might be? Is the load from
cursor passing everything through the coordinating node so I lose
parallel processing? Thanks for any insight.
Pete H