Archives of the TeradataForum
Message Posted: Fri, 12 Jan 2007 @ 14:01:18 GMT
| Subj: | | Re: Table skew and Primary Index |
| |
| From: | | Anomy Anom |
<-- Anonymously Posted: Friday, January 12, 2007 09:02 -->
Doesn't make sense. If the second column alone is not skewed at all, a two column PI that includes that column should not skew either. But skew
may not be critical for small tables (small being relative to number of AMPs on the machine). Even for large tables it may be more important to
choose a useful PI than to be perfectly balanced.
Having a PI that can't be used for queries or joins will usually hurt performance. Storing the data with no skew may make more efficient use of
PermSpace, only to have all your queries skewed at run time. It's always a trade-off. If large fact tables are greatly skewed, though, that's
almost guaranteed to be a bottleneck.
In addition to looking at skew "%", consider the difference between maximum and average space per AMP in terms of bytes (from DBC.TableSize).
That can give you a rough feel of potential performance impact as well as a picture of how much disk space may be wasted due to uneven
distribution.
| 
