Archives of the TeradataForum

Message Posted: Tue, 27 Jul 1999 @ 16:51:29 GMT

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

This is understandable.

Multiload comes into its own whenever it is updating more than one row in a block. ML will only write this block once using specialized Teradata utility code. Bulkload will write the block once for each row inserted into the block using normal Teradata SQL code.

So, as the average number of updates per block increases, ML increases in relative performance over BL. One can calculate this "average" by calculating how many rows per block one has versus the percentage of the table that will be updated. Also, take into account that all rows with the same NUPI will most probably reside in the same block. So, even at a small percentage of the table being updated, if there is a high average number of duplicate NUPIs in the input data, then this will result in a high average hit rate per block updated.

Additionally ML will only write NUSI blocks once no matter how many updates there are for a NUSI value. This gives ML a performance advantage over BL when the table has NUSIs.

Just a note: I believe that support for Bulkload will be dropped soon. This functionality is addressed via TPump, which should have the same performance characteristics. TPump also has row level restart capability which mirrors that of ML.

jdg

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