 |
 |
Archives of the TeradataForum
Message Posted: Thu, 01 Mar 2013 @ 01:39:40 GMT
| Subj: | | Re: Tables: How small is small for TD to dynamically |
| |
| From: | | Attila_Finta |
Ken Hansen wrote:
| | I have a question regarding PI definition for small tables. Teradata replicates small tables across all AMPs (when joined to another table
having a different PI). Therefore we have been recommending that small tables should have a Unique PI that is the same as the PK (to maximize
even distribution). For larger tables I believe we must also consider frequent access points and join paths. Do you have any guidance re: "how
small is small"? How small is small enough for TD to replicate the table? 1 MB? 10 MB? | |
| | When joining two tables with different PIs, Teradata will replicate and redistribute the smaller of the two - irrespective of size but
success will depend upon the size of available spool space. | |
Thanks Ken. Isn't it "replicate OR redistribute"? Aren't those are two very different things? My understanding is that a small code table
will replicated to all AMPs in its entirety (all rows copied to all AMPs) while for a large table (still the smaller one in the join) each of its
joining rows will be redistributed to the single appropriate AMP. So the point of my question is: what's that threshold? Or what am I
misunderstanding?
Attila Finta
| |
