03-03-06 11:50 PM
We experience, as expected, drastic degradation in massload times after
DB2's DIAGLEVEL is set to 4 for additional debugging. You might check
this as it could be the problem.
In another case, we also had unusual behavior in DB2 that somehow
corrupted the load process and caused it to crawl. Specifically,
600,000 address records would load in 15 minutes. Once, something
happened at the 400,000 record and the whole thing slowed down to 1,000
insertions per minute. Needless to say we stopped the run, cleaned up
and restarted. We also ran out of memory on several occasions loading
extremely large xml files which we broke into smaller segments. This
was a 2-tier AIX config running WCBE v5.6.0.4.
From my own experience, the loader package (formerly Catalog Manager)
is not built for speed and reliability. The tools work great for small
loads but I rarely see a customer, or myself for that matter, WOWed by
the utility. If it was top-notch, then Niece wouldn't have made a dime
on Dynaload.
R
If there are IBM'ers listening:
<rant on>Perform better QA on the product with LARGER data sets. I
don't have enough fingers to count all the times when I've found a bug
somewhere because of performance issue related to poor SQL in a command,
not enough indexes, etc. against the WC schema when loaded with __REAL__
customer data. Been a problem since the Net.Commerce days in 1998. It
is getting better...which is a start.</rant on>
> We currently load our inventory data via a csv process that uses the WC utilities
to idresolve and massload the data into the db. We've just upgraded from WCS 5.5 to
5.6.1.1 and DB2 v8 and are seeing a considerable slow down in the massload portion
of
the process. Load times have gone from 2 minutes to about 25 minutes on an AIX server.[vbco
l=seagreen]
>
> Does anyone know of any finetuning we can do to improve performance? Eith
er on the DB2 side or the WC side (ie. Massload configuration). Note, we ar
e currently loading using the "sqlimport" method.
>
> Thanks![/vbcol]
[ Post a follow-up to this message ]
|