Archives of the TeradataForum
Message Posted: Mon, 26 Sep 2005 @ 11:03:24 GMT
| Subj: | | Re: The use of ID Columns |
| |
| From: | | Dieter Noeth |
Naveen Ram Prasanna Kambhoji wrote:
| | How far is it beneficial to define an Identity Column as PI in DWH Applications. | |
In theory a PK (value and definition) should never change, but in reality it does. If you use a dumb artifical PK (doesn't have to be a
sequence) you'll probably never encounter that problem.
But this is about Primary *Key* not Primary *Index*, you should always chosse a PI based on access/joins/distribution.
Btw, in other DBMSes this might be a good idea, because they don't like to join on multiple columns or varchars...
| | I heard that joins to ID columns almost always require row distributions. | |
If the sequence is the PI then not.
Because it's a rumour and rumours tend to be wrong :-)
| | Also what if I want to add more constant values for a multivalued compressed column. | |
create new table -> insert/select -> drop old table -> rename
That's the only way, you'll need twice the perm space (old + new) and enough spool (even if PI is the same and it's a "ins new sel * from
old")
Dieter
| 
