Archives of the TeradataForum
Message Posted: Tue, 12 Feb 2002 @ 09:05:46 GMT
| Subj: || || Re: Insert data into table |
| From: || || Paul Johnson |
I'm with Dieter on this one. I have seen a case where an order item table contains 400m rows with a NUPI of the account number. The
table is very well distributed with about 1% skew. What this hides is that there are about 12 cases where a single account number has of
the order of 100,000 rows. So, although evenly distributed the table contained a small amount of highly non-unique values.
Inserting data was so slow using a SET table that it had to be abandoned. The table was re-created as MULTISET and all is well. The
only downside is the potential for accidentally creating duplicate rows. If you can ensure that this does not happen, or catch it when it
does, then MULTISET is a winner.