Re: Tuning DBWR

Thanks for the suggestion. We determined I/O bottleneck by running glance. Whenever there is a logswitch/checkpoint glance alerts for I/O bottleneck. I did check the alertlog for messages but there was none.

        Changing the db_block_max_dirty_target setting from 50% to 1% resulted in 100% I/O utilization for 30-50 secs every 3-5 minutes, instead of a I/O bottleneck of 2-3 mts every logswitch. During these period we have a very poor response time.

Sandra

In article <947846897.8549.0.nnrp-03.9e984b29_at_news.demon.co.uk>,   "Jonathan Lewis" <jonathan_at_jlcomp.demon.co.uk> wrote:
>
> Not necessarily true in version 8. Multiple
> DBWRs allow for multiple collectors of dirty
> blocks whereas async I/O with one db_writer
> has only one collector even though it has
> many writers.
>
> Also, async I/O allows you to flatten out
> the effects of minor peaks in I/O at a cost
> in CPU and memory. Under heavy I/O load
> async I/O is all cost and no gain - to misquote
> (and mis-attribute) Yogi Bear
> The I/Os are not over until the I/Os are over
> If you have too many I/Os they still have to queue.
>
> Mark Powell has made the most significant point
> so far, I think, regarding the two disc configuration -
> basically this is not enough disks for an Oracle system
> of any size.
>
> Two suggestion
>
> If the problem occurs at log-switches, which probably
> means at checkpoints, this indicates that there are
> very large numbers of dirty blocks which have to be
> written to disk, and DBWR is working like mad to
> dump them. Remove the bottleneck by making
> DBWR work harder between checkpoints. There
> are various ways of doing this, but one is to set the
> (relatively) new parameter:
> db_block_max_dirty_target
> This makes Oracle allow no more than that number
> of dirty blocks in the buffer. It defaults to unlimited;
> try setting it to values like 10%, 20%, 5% of your
> db_block_buffers and see if this improves things.
>
> Alternatively - with apologies for asking potentially
> insulting questions - how are you determining that there
> is an I/O bottleneck ? Is this based on I/O statistics,
> or on the apparent response time from the application ?
> If the latter, then perhaps you don't have an I/O problem,
> you may simply have log files which are too few and too
> small - look in the alert log for messages like:
> cannot switch logfile checkpoint not complete
> when a message like this occurs, no further updates
> to the database can take place until all marked
> dirty blocks have been written, thus releasing old
> redo logs for re-use.
>
> If you have frequent messages of this type then consider
> making your redo logs much larger.
>
> --
>
> Jonathan Lewis
> Yet another Oracle-related web site: http://www.jlcomp.demon.co.uk
>
> ora7dba wrote in message ...
> >What has not been discussed is ASYNC I/O ... using async i/o is
faster than
> >trying to add more DBWR's. HP's are a different animal (compared to
sun's)
> >so you need to check into your OS support for async i/o.
> >
> >-Mike
> >
> ><sandrazil_at_my-deja.com> wrote in message
> news:85iv71$o0f$1_at_nnrp1.deja.com...
> >> Hi All,
> >> I am having a performance problem whenever there is a log
switch, and
> >> when DBWR writes dirty buffer to disk. It takes 2 minutes for DBWR
to
> >> complete its job, during that time there is a I/O Bottleneck. I am
> >> running Oracle 8.05 on HPUX 11.0 ( 32 bit), i am running 4 DBWR's.
I
> >> have performace problem only during this 2 minute interval. Any
> >> Suggestions is highly appreciated,
> >>
> >> Thanks,
> >>
> >> Sandra
> >>
> >>
> >> Sent via Deja.com http://www.deja.com/
> >> Before you buy.
> >
> >
>
>

Sent via Deja.com http://www.deja.com/
Before you buy. Received on Mon Jan 17 2000 - 09:39:44 CST