Archives of the TeradataForum
Message Posted: Tue, 28 Jun 2005 @ 15:59:07 GMT
| Subj: | | Re: Teradata row distribution and hashing |
| |
| From: | | Victor Sokovin |
| | Is there any technique which can help the data architect determine the hash value range if an attribute were added an existing index ?
This can help the architect determine if AMP-local joins will be disturbed by adding or taking out any attribute from an index. | |
I think you already have a good understanding of this, which can be seen in the examples you have provided.
Just a few quick remarks. If you base your hashing on essential business attributes and later want to join on them, then there is not much
flexibility in PI definitions.
Some people use a different approach: they introduce attributes in the PI which are not essential for the business or are even surrogate. This
usually ensures a very even distribution across AMPs but the quality of joins remains to be investigated on a case-by-case basis. Unfortunately,
some sites never start investigation until they run into problems because in the beginning the data model of this kind is likely to perform very
well and only later, when the data acquisition process has matured, the shortcomings become visible.
My feeling is that in general the first approach (business definitions) is preferable to the second one if the project budget does not provide
for the detailed technical analysis of the data model proposed.
Regards,
Victor
| 
