Re: Maximum value for db block buffers

I think I've had one up to 256,000 - but there was so much memory on the box I couldn't think what else to do with it - I couldn't get it any higher because of an O/S bug at the time.

It's an interesting point as to where the bottleneck would eventually have to be.

If you could survive 65535 blocks in the bad old days with only one db_lru_latch, then presumably you could manage at least that number per CPU by setting db_lru_latches = CPU count.

Then there's the checkpoint issue - but that has been improved by the new checkpoint list algorithm.

Multiple db-writers, I/O slaves etc would probably get you round the issue of writing dirty blocks fast enough to keep the LRU clean.

Ultimately, I'd guess that if the SGA had to be be truly enormous because of the size of the application then it would be the redo allocation latch that bottled out before block handling became an issue.

I suspect you would avoid the pre_page_sga option though - otherwise startup might be a bit slow.

--

Jonathan Lewis
Yet another Oracle-related web site: www.jlcomp.demon.co.uk

Filip Hanik wrote in message ...
>I've read earlier in this group people who had values larger than 100,000
>
>Connor McDonald <connor_mcdonald_at_yahoo.com> wrote in message
>news:376CF425.2927_at_yahoo.com...
>> Purely a philosophical question...
>>
>> Given sufficient physical memory - is there a limit on db block buffers
>> where Oracle would start struggling with managing them...?
>>
>> Cheers
Received on Mon Jun 21 1999 - 02:42:18 CDT