Re: Database Hit Ratios

Hi Jonathan,

And anything that makes people think is generally a good thing. That's what I particularly like about Connor's script that adjusts the hit rate to any value you choose. If it makes people re-evaluate what tuning is all about, if it makes them think, "wait a minute, maybe my 99% ain't so good after all !" then that's great. But it's important that people don't then go and jump to some other equally incorrect conclusion.

It's all good fun !!

Richard
"Jonathan Lewis" <jonathan_at_jlcomp.demon.co.uk> wrote in message news:1026116542.14881.1.nnrp-08.9e984b29_at_news.demon.co.uk...
>
> John,
>
> I managed to get a paper of mine in for OW Copenhagen,
> with the title:
>
> THE DATABASE GETS BETTER, BUT THE METRICS LOOK WORSE.
>
> which contains the paragraph:
>
> The ideal was to get a figure close to 100, and the purpose of the
> calculation was simply to answer the question - do I really need
> to buy more memory? However, a frequently overlooked detail of
> the formula was that it assumed that the database was behaving
> as efficiently as possible so that the effects of extra memory would
> simply allow the same amount of 'logical' work to be done with fewer
> accesses to disk.
>
> Of course, this was in the days of Oracle 5.1 when memory was
> very expensive, and designers/developers/coders still worried
> about efficiency at an early stage in development.
>
> Richard is absolutely correct - interpretation, and understanding,
> are crucial. The vicious condemnation, and satirical mocking, of
> the buffer cache hit ratio, in particular, is the result of an urge to
> shock people into thinking. Perhaps it makes it too easy
> for people to believe that "hit ratios are bad", rather than "hit ratios
> are secondary".
>
>
>
>
> --
> Jonathan Lewis
> http://www.jlcomp.demon.co.uk
>
> Next Seminars
> UK July / Sept
> Australia July / August
> Malaysia September
> USA (MI) November
> http://www.jlcomp.demon.co.uk/seminar.html
>
> John Beresniewicz wrote in message ...
> >Richard -
> >
> >Nice job putting the hit ratio "controversy" into a more reasonable frame
> of
> >reference. Your suggestion hit ratios make sense under the assumption of
> >prior minimization of logical I/O's is a good one, but therein also lies
> the
> >"tough nut" to crack. A well-designed system focusing on mimizing
logical
> >I/O for the work to be done is one that has been tuned by design and this
> is
> >of course the best and most cost-effective time for performance tuning.
> >
>
>
>
>
>
Received on Mon Jul 08 2002 - 04:51:58 CDT