Archives of the TeradataForum
Message Posted: Thu, 05 Dec 2002 @ 15:52:42 GMT
| Subj: | | Re: Loading Fact data directly onto Teradata |
| |
| From: | | Judge, James A |
IMO, it depends on the level of "t" in Informatica. If you are doing a lot of aggregation and/or encoding/decoding of dimensional
columns then the Informatica process could be slow; again, it depends. I am seeing use of the "multiple partition" functionality in
Informatica to break up the transaction file and process all pieces in parallel and then merge the result set back together to be passed to
mload process. This has been effective is reducing "t" by 5X is certain cases.
As you stated if you use an "EtLT" approach and load transaction data into a staging area and do most of "T" there you get the benefits
of the parallel database.
A couple of things;
* in Informatica mload script generation that will affect performance; make sure CHECKPOINT value is set to acceptable level
(i.e. not in the '000s of records but rather in minutes or not at all),
* I think Informatica generates a "NULLIF = '*' for each .FIELD in mload script and this can be a host performance issue if
these are not needed - although I'm not sure if that is adjustable without a lot of manual recoding of scripts.
| 
