Re: Any thoughts?

Jim Kennedy wrote:
>
> "Howard J. Rogers" <hjr_at_dizwell.com> wrote in message
> news:4091e982$0$14771$afc38c87_at_news.optusnet.com.au...
> > Sybrand Bakker wrote:
> >
> > > On Fri, 30 Apr 2004 12:49:04 +1000, "Howard J. Rogers"
> > > <hjr_at_dizwell.com> wrote:
> > >
> > >
> > >>Just for the record, Google reports me as being the only person who
> > >>responded to you last time, and that response was a question as to what
> > >>your colleagues at TUSC made of it.
> > >>
> > >>That was hardly a flame.
> > >>
> > >>I agree it was nothing.
> > >>
> > >>But a flame it wasn't.
> > >
> > >
> > >
> > > Which does confirm his TUSC relationships ;-)
> >
> > Oh, the TUSC relationship is irrelevant really, except that I would have
> > thought they would have been the first people to turn to, and their
> > ideas might have been of interest in their own right (which is why I
> > asked the question). It shouldn't stop someone being able to ask a
> > question here, just because they happen to be associated with TUSC, just
> > as it shouldn't if someone works for Oracle (!). On the other hand, he
> > (presumably) has access to some of the more publicised experts in the
> > Oracle publishing world, so one wonders whether they are stumped on the
> > matter, or have even ever been asked about it.
> >
> > What really makes it difficult to provide any answer here is the lack of
> > any substance to the question. I mean, is session logical reads always
> > bigger than consistent gets+db block gets, or always smaller, or does it
> > vary, and if it's one way or the other, what is he doing on the database
> > at the time, and what's the version he's using in the first place, and
> > on what O/S, and so on?
> >
> > And also, why is it important to know anyway? I mean, what's the problem
> > being solved here? What piece of DBA work is stymied because of the
> > inconsistency?
> >
> > I've studied a number of V$ statistics that "sound" like their supposed
> > to add up to some other statistic, but never do. I tend not to lose much
> > sleep over it, unless it poses a quantifiable (and practical) problem,
> > because Oracle internals are, well, Oracle internals.
> >
> > Regards
> > HJR
> >
> >
>
> Howard,
> My understanding (which may be incorrect) is that the V$ stats are not
> beholden to any "database consistency" and are just structures to track
> statistics. In fact, I have run across times when they are very
> inconsistent. (eg wait time writing to a file exceeding any reasonable
> number, due to async IO and the tracking got confused on when the write
> occured.) I am sure there are some subtle things that we may really never
> know about. I agree with your assesment.
> Jim

That...plus bugs. We've got a v8174 database that currently shows hundreds of days of lost time in v$system_event on (say) db file sequential read, which given that I bounced the database yesterday morning....well, you get my drift.

cheers
connor

-- 
-------------------------------
Connor McDonald
http://www.oracledba.co.uk
Co-Author: "Mastering Oracle PL/SQL - Practical Solutions"