Thanks Mark :) I hear your side clear. After being condemned so bad, when I saw that the **10.2** manual was talking about ratios, I thought maybe it was some kind of exception. I know the subject has been beaten to death, but still seeing it in 10.2 manual was what made me ask the question. On 5/16/07, Mark W. Farnham <mwf@xxxxxxxx> wrote:
That depends on your moral matrix. For me, I consider the useless generation of entropy to be evil. Since brain cells firing use energy, thinking about hit ratios generates entropy. Since hit ratios don't amuse me as a puzzle or serve any other intellectually pleasing function and since they serve no practical purpose, I suppose for me consideration of hit ratios is indeed evil. If hit ratios were at least as indicative that something needs attention as an oil pressure light (rather than a gauge) you might look at them. But since assembling the required other information to tell whether the hit ratios mean anything provides more information than the hit ratios themselves and since the hit ratios themselves add no value to the required contextual data to interpret them I find them entirely useless. Now the debate about whether the "first inquiry" should be counted as a hit, a miss, or nothing in formulating the hit ratio equation is mildly entertaining, but since it devolves to the semantic definition of a useless metric that leaves me flat as well. So, yes, at least small "e" evil. The direct questions are: 1) Am I waiting for something that should be in cache at steady state operation that I threw out for lack of allocating memory or slots? 2) Is the wait part of a delay of a process I need to have faster to meet a service level? 3) Is removing the part of the delay caused by the lack of allocation worth the cost to allocate more (whether or not this brings me totally in compliance with the service level) considering relative costs to reduce other parts of the delay? 4) Is it cheaper to over-allocate than it is to get the rest of the performance team to stop worrying about it through logic? (Especially if your machine is memory rich, this is often sadly true). Regards, mwf ------------------------------ *From:* oracle-l-bounce@xxxxxxxxxxxxx [mailto: oracle-l-bounce@xxxxxxxxxxxxx] *On Behalf Of *Ram Raman *Sent:* Wednesday, May 16, 2007 10:16 AM *To:* oracle-l@xxxxxxxxxxxxx *Subject:* Re: Understanding Gets, Pins and Reloads - V$librarycache <snip> I take it using cache hit ratios are not all that evil? <snip>