Re: Declaring Unenforced Constraints

From: Tony Andrews <andrewst_at_onetel.com>
Date: 5 Nov 2004 07:58:43 -0800
Message-ID: <1099670323.302988.198870@f14g2000cwb.googlegroups.com>

Laconic2 wrote:
> So what's the point?
>
> Well, there are a lot of data architects who don't want the DBMS to
enforce
> a constraint "for performance reasons". The merits of that argument
vary
> widely from case to case, but there's some merit to it. It's
particularly
> true if the data can be guaranteed to be "cleansed" by the
transactions that
> write the data.

For a "read-only" database, i.e. a Data Warehouse, this may make sense. For an OLTP environment it really does not. The performance arguments are largely bogus anyway, and performing DIY uniqueness and referential integrity checking in a multi-user database is REALLY HARD, verging on downright impossible, to do right - and is sure to hit performance harder than letting the DBMS do it anyway!

It is especially hard in Oracle, which does not support READ UNCOMMITTED, to prevent 2 users from performing simultaneous transactions that are correct in isolation but corrupt the database when taken together. Received on Fri Nov 05 2004 - 09:58:43 CST