Archives of the TeradataForum
Message Posted: Tue, 19 Mar 2002 @ 23:41:07 GMT
| Subj: || || Re: Data Row Format |
| From: || || Todd A. Walter |
Geoff has this right. There is no reason to group columns together by type. The column order specified in the table should be that
which makes logical sense, not anything related to the physical structure, that is Teradata's job.
re Dave's comment about ordering varchar by access frequency: The difference in performance of accessing varchar columns based on their
position amongst a set of varchar columns will only be noticeable if there are a Large number of varchar (byte) columns. Earlier versions of
Teradata cost more to access columns farther out in the set than current Teradata design requires. Since V2R2, these considerations both for
varchar columns and compressed columns have disappeared.
And finally, while Geoff is correct that we order columns physically basically in their order of appearance within the 3 groups today, we
reserve the right to change this at any time. In fact, on the 64 bit version of Teradata where word and doubleword boundaries make a big
performance difference, the physical ordering will be quite different - based on an algorithm that maximizes the placement of appropriate
sized fields on appropriate byte boundaries for performance. All done internal to Teradata of course, you specify the logical ordering and
we'll figure out how to layout the rows.