Archives of the TeradataForum

Message Posted: Tue, 30 Mar 2004 @ 14:46:47 GMT

	<Prev	Next>		<<First	<Prev	Next>	Last>>

Three things strike me:

1) ETL tools do not absolve you from understanding the attributes of the load method. Each tool has a documented set of strengths and limitations. I'd start the topic as "programmer proficiency using ETL tools" then move to Teradata or ETL tool performance.

2) We have generally experienced excellent throughput with Informatica 6.1 & Teradata V2R5 because we carefully consider the above item. The performance is quite predictable; when it falls outside of the predicted range we know we need to look at something like the addition of an index to support a lookup, etc.

3) The Teradata Solaris ODBC driver reminds me of a circa 1996 ODBC driver - ODBC moved from an idea to a viable approach then because the specification provided for an improved method of bulk loading. The principle evidence that Solaris ODBC is inefficient is that we average 10K rows/minute using ODBC and about 100K rows/minute using Tpump. Since both ODBC and Tpump are statement based loading methods we should not see such a wide divergence of performance. A well written ODBC interface should be within 90% of the performance of Tpump.

ETL tools provide increased functionality that is completely dependent on a quality ODBC interface. In our environment we'd love to use Informatica's constraint based loading feature more often, but the pathetic ODBC driver makes it impossible. Constraint based loading means the tool resolves loading sequence requirements imposed by referential integrity - handy and exactly why you want to use an ETL tool.

	<Prev	Next>		<<First	<Prev	Next>	Last>>