Re: Challenge: Partitioning is a wrong idea

fitzjarr..._at_cox.net wrote:
> The first example, of deleting a years'
> worth of data out of a 1 terabyte table containing 8 years of records
> is an excellent choice, and you summarily discounted the response, as
> should be expected from someone trolling a newsgroup to fan flames.
A
> simple alter table ... drop partition ... would be far faster and
> generate much less redo and much less rollback/undo than a delete
from
> sales_history where txn_date <= ..., and would consume less cpu in
the
> process.

So deleting one year of data, is it logical or physical operation? It can't be logical, because the decision when to delete and how much to delete is completely arbitrary. Users most likely don't care about the old data, this is why DBA has a certain freedom when to delete data, and how to delete it. Now, if users are isolated from this deletion transaction, why not run it as a slow background process? Why all this rush?

> You'll discount this respones, as well, as it didn't offer any
> 'scientific proof' you so adamantly demand. The truth is table
> partitioning is not a cut-and-dried scientific operation, it's a
> subjective task best left to those who understand its nature.
> Partitioning is not for everyone, nor for every situation. Don't try
> to disuade people simiply because your closed mind can't grasp the
> benefits partiioning can provide.

Finally, the truth surfaces. So, unlike the other, really useful features, you can't really defend partitioning ("it's not for everybody","your mileage may vary","no silver bullet"). Not that I believe that with your style of argument you were able to prove something really obvious, like benefits of B-tree indexes. Received on Thu Apr 07 2005 - 17:54:02 CDT