Archives of the TeradataForum

Message Posted: Tue, 01 Nov 2005 @ 12:57:40 GMT

	<Prev	Next>		<<First	<Prev	Next>	Last>>

The thread has been developing very interestingly so far by revealing different definitions and calculation techniques for the "data skew". "Data skew" is one of the most popular subjects in Teradata and yet at the same time one of the least understood subjects among the Teradata users. I think one of the reasons for confusion is the fact that "skew" is a very generic mathematical property of random distributions. It is applicable in many different situations and it does depend on the context.

I like the definitions I have found in the "Statement and Transaction Processing" manual:

www.teradataforum.com/teradata/20050610_204651.htm .

Unfortunately, not all the manuals are clear about the definitions they use. Another source of confusion.

Even in this thread folks have been talking about the attribute value and partition skews. The former is the most generic, it has nothing to do with Teradata as such. The latter has a lot to do with Teradata and even the particular instance of the database and the table in question: number of AMPs, hash algorithm, PI chosen, etc.

BTW, if somebody missed out on the thread in summer, please feel free to comment on it as well. I think there is still more to say on the "data skew" in the context of stats collection.

Regards,

Victor

	<Prev	Next>		<<First	<Prev	Next>	Last>>