Re: indexes vs sorted files in range and equality search

From: <>
Date: Thu, 6 Dec 2007 08:45:26 -0800 (PST)
Message-ID: <>

On 6 Dic, 17:21, paul c <> wrote:
> wrote:
> ...
> > is my analisys basically cannot-deny-that-so-oh-my-god-accomplished-
> > correct? :)
> For those conclusions, it depends on the data, I'd say you need more
> assumptions, eg., whether the result sets are singletons, whether joins
> are not involved, whether some columns are very few-valued, also maybe
> why a zero-time hash is possible for the data. Might be better to try
> to characterize the cost for each returned "record", then it might be
> clearer what the total cost of the query is.

thank you for answering. so, you're saying that, for instance, the effective choice of sorted files over unclustered indexes in range search operations depends on wheter or not the result sets will be singleton. if so, than there'll be no need to choose sorted files, otherwise, i'll avoid unclustered indexes because of the possibly massive I/O operations number. well i could imagine that in a real life db scenarios, no 100% sure conclusion can be taken, since terms in the mathematical analysis can change greatly, depending on the data. so my question could shift into another one: do in common scenarios equality searches on a field usually return resultset so large that sorted files are better than unclustered indexed ones? well, to be honest, i don't expect anyone answers me this statisticaloriented  question in this .theory group, however let me thank you to confirm me my initial (although not shown in the first post) thought: theorical analysis can seldom be useful a priori, without pointing out the data instance the question refers to.

anyone has other ideas about? Received on Thu Dec 06 2007 - 17:45:26 CET

Original text of this message