Re: very low hit ratio

"Carl Kayser" <kayser_c_at_bls.gov> wrote in message news:c5ol4r$b39$1_at_blsnews.bls.gov...
> I don't know about DW (and I'm primarily a Sybase DBA) but 90% hit ratio
> means a 10% miss ratio which I do not consider to be good. Consider
chapter
> 7 of Alan Packers book which is available (along with the TOC) online at
>
>

http://www.amazon.com/gp/reader/0130834173/ref=sib_dp_rdr/104-8369762-3779959#reader-page
> >

I'm not convinced - but then since I consider hit-ratios to be misleading perhaps I am biased.

This chapter talks about not setting a target, about concentrating on the miss ratio to reduce physical reads. However at no point does it consider whether the workload is optimal, nor does it mention - except in passing - the end-user response time. If halving the miss ratio from 10% to 5% results in a 0.05% change in response time what is the point. What happens if halving the miss rate results in much *increased* response time - inappropriately choosing indexed access paths will frequently do this.

It also begs the question what do you consider to be 'good' and why.

By contrast a response time focussed approach can ask of the end-users, or the business management, questions like how long should it take to enter an order/how many orders should we process per day, what should our batch window be in hours. Much more productive.

-- 
Niall Litchfield
Oracle DBA
Audit Commission UK
http://www.niall.litchfield.dial.pipex.com/