Re: cache buffer chain latch

cache_buffers_chain

    used to default to next prime after db_block_buffers/4

can be set using parameter

    _db_block_hash_latches

Each database block is associated with one buffer chain. When a block is required, Oracle generates a hash value for that block to determine which chaing it should be on and checks this hash chain to see if it is already there. If it is, fine, if not Oracle loads it into a buffer and attaches the buffer to the chain. For stability reasons Oracle can take the latch for a hash chain at various times during the life cycle of a block request.

One reason for lots of waits is that a lot of frequently used blocks are associated with the same latch - look at v$latch_children instead of v$latch for symptoms of this. Check X$BH for the HLADDR column

    select hladdr, count(*) from x$bh
    group by hladdr
    having count(*) > 10 (say)
Select

    file#, dbablk from x$bh for those hladdr values.

Another, more likely, reason is that your slowdown is actually caused by heavy
checkpointing.

--

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

Allan Plesniarski wrote in message <85nk9m$47v$1_at_nnrp1.deja.com>...
>Hello
>
>I am having a problem, every couple of minutes the cpu stops running at
>99% and drops to under 50% utilization. This lasts for a minute before
>the system picks up again and continues merrily on it's way.
>
>The wait that is blocking the processes is a "cache buffer chain" latch
>(different to cache buffer lru chain).
>
>Unfortunately I have very little information on the cache buffer chain
>latch. The only thing I found was a vague reference to freelists but
>the process is not doing any large amount of inserting at that time.
>The processes are just doing queries at that time, no commits/inserts
>or deletes are going on.
>
>If anyone knows of some references, it would be appreciated.
>
>Thanks
>
>
>Sent via Deja.com http://www.deja.com/
>Before you buy.
Received on Sat Jan 15 2000 - 14:15:55 CST