Re: cache buffers chains latch

DB block buffers are associated with hash buckets, and the buffers for a single bucket are in a linked list (the chain). Typically you would expect about 4 buffers per hash-bucket.

This latch is actually the parent latch for the 'db_block_buffers/4 plus a bit' latches. When Oracle needs to examine a block, it works out which hash bucket it ought to be in, then grabs the latch for that hash bucket before running the linked list.

You figure for misses is a little high, but not extremely so. You could check v$latch_children to see if there are just a few of the latches which are responsible for most of the misses, then check x$bh to find out if there are any significant blocks on those latches which could be kept away from each other (i.e. in different buckets) by changing the value of db_block_hash_buckets.

I think I wrote this up for my web-site at some point - if not check dejanews for some messages I posted about 4 months ago.

NB DB_BLOCK_HASH_BUCKETS defaults to the next prime number above db_block_buffers - so there will be lots of child latches.

--

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

NetComrade wrote in message <37d960d2.278121933_at_news.earthlink.net>...
>cache buffers chains
>Gets Misses sleeps
>8571094116 59012873 225223 56053565 23
>
>Some stuff I got off metalink
>
>Cache buffers chain latch:
>..........................
> This latch is related to DB_BLOCK_HASH_BUCKETS (undocumented in
> Oracle7) which is related to db_block_buffers. It can be tuned by
> adjusting DB_BLOCK_BUFFERS.
>
>but it doesn't tell me anything! Neither does it explain what the hell
>this latch means.
>---------------
>Andrey Dmitriev eFax: (978) 383-5892 Daytime: (917) 373-5417
>AOL: NetComrade ICQ: 11340726 remove NSPAM to email
>
Received on Fri Sep 10 1999 - 17:46:40 CDT