|
|
Archives of the TeradataForum
Message Posted: Mon, 14 Aug 2006 @ 22:11:31 GMT
Subj: | | Re: Alternative to Insert/Select |
|
From: | | Stover, Terry |
One thing to remember with multiload is that it is incompatible with join indexes, hash indexes, USI's and identity columns. JI's and
aggregate JI's are really useful, especially with very large tables.
You mention the target table is 1.3 Terabytes and the nightly data is 6-8 million rows, so we don't really know the relative size of the two
data sets. How many rows is the target table and how many bytes is the nightly data?
I would question is the need to merge the data by doing an insert select into an empty table. You may be better off doing the insert select
into the target table, rather than merging the work table and target table into a new empty table. I can't imagine the cost of the TJ for the
insert select is going to be more than the cost of copying 1.3 Terabytes and recollecting stats. I vaguely recall reading that multiload is
generally faster than a fastload - insert/select when a significant portion of the data is affected (10% comes to mind). As with all things
Teradata it depends on your data, but it's worth benchmarking. I've been in situations where the chosen load method was benchmarked with a
limited data set (1 year of data). When you get to a broader data set (5 years data), the benchmarks favor a different solution.
| |