> >Hmmm.. from "Oracle Performance Tuning 2nd edition" (Gurry and=20 >Corrigan,=20 >O'Reilly Press, 1996) -- so were talking version 7 RBO here: > >"Contrary to popular belief, COUNT(*) is faster than COUNT(1). If the=20 >rows are being returned via an index, counting the index column--for=20 >example, COUNT(EMP_NO) is faster still. [followed with actual=20 >test runs=20 >including timings to prove this]" > >So, for the sake of historical accuracy (and not much else), there is=20 >some evidence that a long time ago the count(*) performance=20 >issue existed. > >-Mark Bole >
What is interesting is that it is contrasted with a select count() against a column on which a unique index exists - which is shown to be quicker still. Which suggests that (at that time at least) the optimizer was not smart enough to find and use a unique index without lots of help.
I also remember about that time finding advice on Metalink to do a count(*) on a table if one was getting ORA-01555 (See ORA-01555.ora-code.com)'s due to delayed block cleanout - this was supposed to be the 'best' way to force every block to be read. -- http://www.freelists.org/webpage/oracle-l
