Archives of the TeradataForum
Message Posted: Tue, 30 Nov 2004 @ 21:43:03 GMT
As you know, dates are stored internally as integers. Also, a series of consecutive dates contains gaps if we view the dates as integers. For instance, the sequence of "consecutive" dates 2003-12-30 through 2004-01-02 is actually the sequence
1031230 1031231 /* gap here */ 1040101 1040102
and, of course, there are similar gaps elsewhere between months.
Now, if we use a date as the primary index, the distribution turns out to be less even than we might like. I defined a table with a date as the primary index and stored twenty years' worth of consecutive dates in it. The ratio max(CurrentPerm)/avg(CurrentPerm) is near 1.0 for our better tables, but for this table it is 1.26.
Now my question. Is this simply an unavoidable consequence of the gaps in the integer sequence, or might there be some way to make the hashing of dates as even as that of integers?
|Copyright 2016 - All Rights Reserved|
|Last Modified: 23 Jun 2019|