Archives of the TeradataForum

Message Posted: Thu, 27 Sep 2001 @ 19:25:20 GMT

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

I'm looking for some tuning advice for Multiload. We're running a 3 node 4800. I have an historical table, by week, that currently has 14 weeks of data or 1.5 billion rows. 110 million rows are being added to it. It needs to grow to be at least 25 weeks of data or 2.5 billion rows (users want a years worth! eek!)

Here's my problem:

Application built this table with 3 NUSIs on it and started loading. They're telling me that the load of 110 million rows is taking over 14 hours. To get the back load run (we're trying to backfill the 25 weeks), I've suggested dropping the NUSI, getting all the loads run, and then rebuilding the indexes and collecting stats.

This will help in the short term loads, but I don't know that we can incur the time to rebuild the indexes and collect statistics every week. Does anyone have a rule of thumb of the break-even point of dropping NUSI's during multiload? As we get more and more weeks of data loaded, dropping and re-creating the NUSI's will not be a viable option (I'm guessing). I thought about suggesting a fastload into an empty table and then doing a muli-step insert to combine the tables, but persume that I'd have problems with a) space and b) time that takes.

Any suggestions?

Ruth Fenwick
Hudson's Bay Company

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