Re: Design question-versioning using timestamps on rows

nospam_at_satan.com (Darby Crash ) wrote:
>
>I'm not sure what all you gain by having the timestamp in the
>primary key and not as a non-key mandatory attribute. You toss
>normalization because now the primary key is no longer minimal.
>Also, as you have pointed out you probably will take some hits
>performance-wise as a result of the design. I think I would look for
>some object solution from Oracle 8 or write the joins explicitly with
>the key and the non-key version attribute.
>I would also be wary of modeling based upon commercial software. You
>need to remember that their databases are dedicated to that applicatio
>n.
>Any change is very controlled.
>
[snip]
Since the original poster indicated that he would not be time-phasing the entries, then your normalization comment applied, at least, to the logical PK, e.g. that which uniquely identifies all other attributes to the extent that other determinates are candidate keys.

In most RDBMSs, creating a unique index including the time/date stamp would increase performance, this is called _coverage_. This redundancy is _automatically_ maintained by the DBMS and does not impact entity or referential integrity. However, there might be an update penalty depending on the DBMS and transaction volume(s).

-- 
*****************************************************************************
* K. E. Fein                      "To thy own opinions be true...           *
* kef_at_tucsonpo.jewelway.com         ...those expressed herein are but mine" *
*****************************************************************************