Re: 10g RAC and db_multi_block_read_count
Date: Wed, 5 Aug 2009 09:08:33 -0700
Message-ID: <a9c093440908050908h501f6ba7t6e61a5d15c011e16_at_mail.gmail.com>
Could you cite the reference that you mentioned about the MBRC=8 on RAC suggestion?
What is your workload? DSS? OLTP?
Given that not setting db_multi_block_read_count basically results in a costing of 8 w/o system stats but I/O size is db_multi_block_read_count blocks or in this case 1MB.
I look at it like this: is it more efficient to execute a larger number of smaller I/Os or a smaller number of larger I/Os when doing a FTS operation (the answer is the latter)? Then I would ask myself, does it even make any difference if it is RAC or not, and if so, why? I am struggling to think of any significant difference that RAC would introduce, so I would comment that it makes no difference and if the comments you found have no technical explanation, I would classify it as mis-information.
On Wed, Aug 5, 2009 at 8:28 AM, Crisler, Jon<Jon.Crisler_at_usi.com> wrote:
> If I allow Oracle 10g to automatically set the parameter
> db_multi_block_read_count, the value is determined to be 128 (8k
> blocksize). However, I ran across some comments that suggest in a RAC
> environment that this large size might be a bad choice, and it should be
> backed down to 8. I cannot find any info in statspack / AWR that
> suggests a smaller number is better in RAC.
>
> Does anybody have further insight into this value in a RAC environment ?
-- Regards, Greg Rahn http://structureddata.org -- http://www.freelists.org/webpage/oracle-lReceived on Wed Aug 05 2009 - 11:08:33 CDT