Re: db_block_lru_latches

Compare your statement with mine:

    "can" appear
    "redundant" buffers
    "seem"
    "can result"
    "high-speed OLTP systems"

compared with:

    "we are using all buffers "

I think you should note that my observation is not a generic, or absolute, claim, and identifies two implementation requirements for the problem to appear.

Your comments describes a system which fails to meet the specification on one of the stated requirements.

Of course I agree with your comment about "bounding", For the record, the most recent system I saw this on was 8.1.7.4 running on 64-bit HP-UX - but it should affect all versions from 7 onwards.

--
Regards

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

Next Seminars
 UK  Sept, Nov
 USA x 2  November

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

Mike Ault wrote in message
<37fab3ab.0208091023.42a81aef_at_posting.google.com>...


>It would nice for you to bound your statements Gentlemen as in "At

>least with Oracle8.1.x.x on such and such an OS I have found that

>using greater than x db_block_buffers..." you make it sound like all

>versions of Oracle are affected by these problems.

>

>At my current client we are using 400,000 16K buffers and have not

>seen the problems you are stating on 8.1.7.3 64bit on HP. In fact, we

>are using all buffers most of the time and have not seen any of the

>buffer problems you describe.

>

>Mike Ault

>

>"Jonathan Lewis" <jonathan_at_jlcomp.demon.co.uk> wrote in message

news:<1028878742.7095.0.nnrp-01.9e984b29_at_news.demon.co.uk>...


>> There is at least one problem that can appear if you

>> set db_block_buffers too large -  redundant buffers seem

>> to allow Oracle to breach the _max_cr_dba_block

>> limit in high-speed OLTP systems.  This can result

>> in very long chains of CR blocks for critical blocks,

>> which results in the associated CBC latch being

>> held for a long time and causing silly amounts of

>> contention.

>>