Archives of the TeradataForum
Message Posted: Wed, 23 Feb 2005 @ 10:05:25 GMT
| Subj: || || Re: Optimal PI |
| From: || || Victor Sokovin |
| ||On picking a PI, how bad does distribution (skewness) need to get before one abandons an optimal access PI. Also, how bad does a volatile
PI need to get. Let's assume duplicate row checking is not a factor (mulitset or USI defined). Any rule of thumb or advise, appreciated.|| |
I am afraid I cannot provide a universally useful advise or a rule of thumb but, in my opinion, the right PI can be found only by experimenting
with a large and representative sample of real data. It is a matter of balancing the ETL processes, typical aggregation jobs and end-user queries.
You might need to foresee them if you are in the early phase of the project. One should also take into consideration the other tables (and their
PI) to which the table in question is going to be joined.
I would rather invest the time into this type of experiments than spending it on reading theoretical recommendations. In my experience, it is
better to read them after you have set up a few tables by yourself.