Re: Insert contention on RAC
Date: Thu, 26 Jun 2008 16:46:15 +0200
It used to be like this the the freelists number * 5 = number of blocks allocated when bumping the HWM. However in OPS it would be capped to the number of blocks that were covered BL locks (assigned with gc_* parameters). But the fact you see a buch of HW enqueue contention tells me the the HWM is not bumped with 5*nr freelist blocks. In fact now with dynamic BL locking there should be a 1 to 1 relateion between lock and block. Dump some redo and you can see the bumping of the HWM with the number of blocks. I assume it is 1 or 5 and not the high number that you want.
On 6/25/08, Baumgartel, Paul <paul.baumgartel_at_credit-suisse.com> wrote:
> To borrow the style from some MetaLink notes--
> Fact: Oracle 10.2.0.1.0
> Fact: RedHat 3
> Fact: 3 node RAC
> Many concurrent sessions inserting one row at a time into the same table.
> On first run, table in ASSM tablespace, about 50% of database service time
> was gc buffer busy waits. Built a new tablespace with manual segment
> management, re-created table with three freelist groups and 4 freelists.
> Next run showed >90% of service time was enq: HW contention; total run time
> was slightly higher than it was with ASSM.. Re-created table again,
> allocated three 4 GB extents, one for each instance. Current run is still
> showing high HW contention waits.
> Might I have too many freelists? Other than forcing all sessions to use
> the same RAC instance, is there anything else I can do to reduce these
> *Paul Baumgartel*
> *CREDIT SUISSE*
> Information Technology
> Prime Services Databases Americas
> One Madison Avenue
> New York, NY 10010
> Phone 212.538.1143
> Please access the attached hyperlink for an important electronic communications disclaimer:
-- Anjo Kolk Owner and Founder OraPerf Projects tel: +31-577-712000 mob: +31-6-55340888 -- http://www.freelists.org/webpage/oracle-lReceived on Thu Jun 26 2008 - 09:46:15 CDT