Hi All,
We are getting large waits for 'db file sequential
read' and 'db file scattered read' on the production
system. How has your individual experience been on
changing (increase/decrease) the db_block_buffers and
db_file_multiblock_read_count in such cases ?
Thanks,
Deepak
- "Gait, Christopher" <cgait_at_condor.nrl.navy.mil>
wrote:
> Jack,
>
> I have a few comments on places to look. A lot of
> the items near the top of
> your list are normally ignored (like the 'SQL*Net
> message from client') but
> very high figures can indicate a networking issue
> either in how the network
> itself is doing (competition from other traffic,
> packet size, presence of a
> 'chattering' card on the net, etc.) or a Net8 issue.
> The high number of db
> file scattered reads may be of more concern,
> however, since it shows a lot
> of full table scans are going on. This would point
> to getting on your
> developers 'top ten SQL queries from hell' and
> optimizing them. This usually
> represents a large part of the problem that is
> 'Oracle's fault' or 'that
> dang DBA.' Be nice to your developers, as a rule,
> but if they continue to
> write badly tuned SQL that ignores indexes and just
> look at you oddly when
> you talk about cost-based optimization and hints,
> consider having one of
> them hung in a cage above the cubicles as an example
> to the others.
>
> I would be interested in seeing the results of this
> query on your system at
> a busy time:
>
> SELECT
> Event,
> SUM(Total_Waits),
> SUM(Total_Timeouts)
> FROM
> V$Session_Event
> WHERE
> total_timeouts > 0
> GROUP BY
> Event
>
> This should pop some problem children to the
> surface. Some of your values
> look like you have a similar problem to us: bad I/O,
> particularly redo. We
> have always had problems with redo, and could still
> use more tuning.
>
> How balanced is your I/O? Do you have the luxury of
> having things spread
> out, or are you stuck with what an SA laid down in
> concrete for disk/array
> setup? Also, what kind of a system is it, OLTP, DSS,
> mixed? High transaction
> rate?
>
> I just read in your next message in this thread that
> you moved to a 4K block
> size. Generally not a good idea, though I suppose
> there are some really
> high-transaction OLTP systems that can actually
> benefit. We're comfortable
> with an 8K on both our OLTP and DSS instances, but
> then we have a
> transaction rate a fast typist could probably keep
> up with (~1,000
> transactions a day) and you should probably consider
> 16 or higher for
> anything faintly resembling a warehouse.
>
> Regards,
> Chris Gait
>
> -----Original Message-----
>
>
> Hi All,
>
>
> I have run utlestat & utlbstat plus some other
> scripts and have a lot of
> wait
> events
>
> *************************
> Event Name Count Total
> Time Avg Time
> -------------------------------- -------------
> ------------- -------------
> SQL*Net message from client 443489
> 4645139 10.47
> rdbms ipc message 1995
> 999996 501.25
> db file scattered read 1087432
> 798071 .73
> PL/SQL lock timer 1
> 10001 10001
> buffer busy waits 3542
> 2415 .68
> latch free 6804
> 2149 .32
> db file sequential read 35326
> 1812 .05
> SQL*Net message to client 443501
> 617 0
> log file sync 1354
> 514 .38
> SQL*Net more data to client 5142
> 98 .02
> control file sequential read 18
> 4 .22
> enqueue 1
> 0 0
> file open 13
> 0 0
> refresh controlfile command 6
> 0 0
> 14 rows selected.
> SVRMGR>
> SVRMGR>
> SVRMGR> Rem System wide wait events for background
> processes (PMON, SMON,
> etc)
> SVRMGR> select n1.event "Event Name",
> 2> n1.event_count "Count",
> 3> n1.time_waited "Total Time",
> 4> round(n1.time_waited/n1.event_count, 2)
> "Avg Time"
> 5> from stats$bck_event n1
> 6> where n1.event_count > 0
> 7> order by n1.time_waited desc;
> Event Name Count Total
> Time Avg Time
> -------------------------------- -------------
> ------------- -------------
> rdbms ipc message 5441
> 1383003 254.18
> smon timer 7
> 210005 30000.71
> pmon timer 668
> 200693 300.44
> log file parallel write 1456
> 499 .34
> control file parallel write 666
> 420 .63
> latch free 8
> 5 .63
> control file sequential read 12
> 0 0
>
>
> *************************
> The most of the buffer busy waits were on data
> blocks belonging to one table
> only.
> The point is that performance of som infrastructure
> testing has decreased
> dramatically between two releases of this new soon
> to be live application.
> One of the requirements of the suppliers was to
> switch from 4k to 8K block
> size.
> so I rebuild the database and left all other parms
> the same.
>
> Now performance is poor and they all point the
> finger at ORACLE. I defend
> oracle
> and say it's the application, but in the meantime
> will be rebuilding the
> database to 4K. Does anybody see anything else I
> should be looking at ?
>
>
> Jack
>
>
>
>
> De informatie verzonden met dit E-mail bericht is
> uitsluitend bestemd voor
> de geadresseerde. Gebruik van deze informatie door
> anderen dan de
> geadresseerde is verboden. Openbaarmaking,
> vermenigvuldiging, verspreiding
> en/of verstrekking van deze informatie aan derden is
> niet toegestaan.
> Ernst & Young staat niet in voor de juiste en
> volledige overbrenging van de
> inhoud van een verzonden E-mail, noch voor tijdige
> ontvangst daarvan.
>
> The information contained in this communication is
> confidential and may be
> legally privileged. It is intended solely for the
> use of the individual or
> entity to whom it is addressed and others authorised
Received on Tue May 16 2000 - 17:28:34 CDT