Re: Design question : Normalize or Not to normalize ?

On 2004-10-14, Tony <andrewst_at_onetel.com> wrote:
> ntareen_at_yahoo.com (Nisar Tareen) wrote in message news:<6134f1ca.0410131241.2d3cc4cc_at_posting.google.com>...
[deletia]
> It is not merely "academic" to normalise, it is "practical" too.
> Denormalised designs require more complicated code to keep the
> redundant data in sync with reality, and are prone to errors. In my
> experience, when someone makes a denormalisation like this in an OLTP
> database, 9 times out of 10 they do it without any real justification
> other than "well, I think it will help performance". They should
> concentrate on sorting out the real performance issues, not
> denormalising based on knee-jerk "practicality". In 14 years or so I
> have never (and I do mean NEVER) come across a case where a
> denormalisation of this kind was really justified. The trouble is,
> everyone knows how to denormalise; real performance tuning takes more
> skill.

So? How do you go about tuning the performance of a query that requires a 3-5 way join due to normalization? What about the case where the 5-way join would tend to generate full scans on multiple tables where the denormalized version would be index range scans?

The 2-way joins are just the starter.

-- 
        Negligence will never equal intent, no matter how you 
attempt to distort reality to do so. This is what separates         ||| 
the real butchers from average Joes (or Fritzes) caught up in      / | \
events not in their control.