Date: Tue, 29 Jul 2008 09:08:30 -0400
"David BL" <davidbl_at_iinet.net.au> wrote in message
> On Jul 29, 1:18 pm, "Brian Selzer" <br..._at_selzer-software.com> wrote:
> > "David BL" <davi..._at_iinet.net.au> wrote in message
> > news:f08bba46-7bd1-443e-97fd-fe07a9ec1a3f_at_25g2000hsx.googlegroups.com...
> > > On Jul 29, 10:45 am, "Brian Selzer" <br..._at_selzer-software.com> wrote:
> > > > "David BL" <davi..._at_iinet.net.au> wrote in message
> > > > > Physical duplication of data can be appropriate to increase read
> > > > > performance at the expense of write performance. Indeed any
> > > > > secondary
> > > > > index is a form of redundancy that hurts write performance.
> > > > Boosting read performance can be accomplished just as well with a
> > > > covering
> > > > non-clustered index as with a clustered index.
> > > That is not always true. There could be an application involving a
> > > query that uses the non-clustered index and also needs *all* the
> > > additional data in the record. The additional seeks could mean the
> > > read performance doesn’t meet the requirements.
> > If the index is a /covering/ index, then there is no need for the
> > additional
> > read.
> But then it won’t be a non-clustered index.
Yes it would. The additional read and an exclusive lock /would/ be required whenever there is an update. You appear to be equating an index that covers the heading with a clustered index. Covering the heading--that is, containing all of the columns in the heading--is not what makes an index a clustered index. Received on Tue Jul 29 2008 - 15:08:30 CEST