Re: Is it better to set the freelist to a prime number ?

Now there's an interesting point.

Unless I missed something, I was thinking that the PID referenced was v$process.pid (the Oracle process id) whereas you seem to be suggesting that it is v$process.spid (the operating system process id). Do you want to do the test, or shall I ? I have a free slot some time next March ;)

(Note too, that your argument would only require an ODD number of freelists, not
a prime number).

I agree with your comment about pattern
avoidance, though.

--
Jonathan Lewis
http://www.jlcomp.demon.co.uk

Next Seminars
 UK  Sept
 Australia August
 Malaysia September
 USA x 2  November

http://www.jlcomp.demon.co.uk/seminar.html

Andrew Mobbs wrote in message ...


>Jonathan Lewis <jonathan_at_jlcomp.demon.co.uk> wrote:

>>

>>Thanks for the reference, which I've followed up.

>>Suddenly it's all clear.

>>

>>It makes no difference whether the value is prime

>>or not.  The significance appears I think when you

>>use multiple free lists AND multiple freelist groups

>>in a single instance.

>>

>>In this case, the number of freelists and the number

>>of freelist groups must be relatively prime (co-prime ?)

>>or you don't make use of all the freelists in all the

>>freelist blocks.

>

>Not quite, it still can make a difference in some cases. For example,

>if there's some characteristic of your environment that makes the PID

>non-random.

>

>Say I start up a batch job that runs as 64 parallel processes on the same

>system as the database server is running.  Each batch job has an Oracle

>shadow process. It's quite likely (on some OSes) to have the application

>and shadow processes interleaved, i.e. all the application processes to

>end up with an odd PID, and all the Oracle shadow processes to end up

>with an even PID (or vice-versa). There's an example of something very

>similar on my other terminal about two feet away :-)

>

>If you had a even number of freelists, you'd only end up using half

>the freelists.

>

>PIDs have no guarantees about their numerical properties, so may well

>fall into a pattern (possibly one more complex than just all being

>even). Primes have the greatest resistance to matching a pattern.

>

>--

>Andrew Mobbs - http://www.chiark.greenend.org.uk/~andrewm/