Re: Re: larger block sizes
Quote:
Originally posted by TomKyte
since temp is read just using multi-block IO's like a full scan? It reads N kb of information, it does not read a block (so, it boils down to 6 one way and 1/2 dozen the other, we read 128k -- or Nk in general)
And also, since TEMP does not do LIO at all, what is the comment about LIO? what does that mean exactly.
See this posting for the entire discussion.
Don't hold your breath ... I asked Mike about this on Mar 8th http://dba.ipbhost.com/index.php?showtopic=1239&st=30
Re: Re: Re: larger block sizes
Watch out or he'll take you to task
Based on that I'm sure he'll be back to answer.
Re: Re: larger block sizes
Quote:
Originally posted by slimdave
Mike, this comment intrigues me, as it implies to me that you believe that a test on a 32bit version of Oracle would not be valid, or something?
it is that you cannot create 32k blocks on 32 bit OS's (I cannot :(
but it does beg the question how he knows it'll show a 1/3 increase in performance but that is another story. (usually you run a benchmark to discover the results, not announce the results you'll BE benchmarking in the future? that is how it has worked for me in the past, that the results were found after the benchmark was run)
Once upon a time ago, these 32k blocks had a special name...
BOB - the Big Oracle Block, it was for the incredible addressing space of 64bit.
if you google for
"big oracle block"
you'll get one hit (like magic, until and if this page gets indexed).
We used to have an option call the VLM option (Very Large Memory) for this support -- not around anymore, but you need 64bit to get 32k blocks.
I think though that Microsoft came out with MS Bob and we backed off. Something like that.