RE: Rman io done waits and the rate parameter
- From: Fuad Arshad <fuadar@xxxxxxxxx>
- To: DENNIS WILLIAMS <DWILLIAMS@xxxxxxxxxxxxx>, Oracle-L <oracle-l@xxxxxxxxxxxxx>
- Date: Mon, 20 Dec 2004 11:05:44 -0800 (PST)
Deninis,
i agree with you
the interval i took the snapshot was between the time the backups werre. running
we are not seeing i/o at the disk subsystem level.
the disk cache is being 100% utilized .
the problem here is the system is a very busy critical system
a transaction taht takes 1.2 seconds during peak volumes without rman running
goes upto 20 secs
the problem here is to determine are we throwing too much at the tapes that is
causing these excessive waits.
We are using async i/o and even though our volumes are not very evenly
configured.
i am looking at ways to manage backups and not interfere wilth the the
application
the problem here is thisis a 24/7 health care application where 3rd party
transactions occur every minute of the day.
DENNIS WILLIAMS <DWILLIAMS@xxxxxxxxxxxxx> wrote:
Fuad
Statspack summarizes waits in two dimensions - over the time period
between snapshots, across all processes on the server. So be careful about
drawing conclusions from Statspack reports.
I don't use RMAN to write to tape. However, in reading the RMAN
documentation, it appears the main problem is keeping the tape buffer filled
so the tape can keep moving. Apparently a tape drive gets really slow if the
tape has to stop and restart. Most tuning is oriented toward getting more
I/O flowing to the tape drive. So if you are trying to slow down the amount
of data flowing to the tape, you may end up with greater problems.
When you say application timings go up by 20% when RMAN is running, I'm
assuming you are saying that the RMAN activities seem to be slowing the
other applications down.
Perhaps you can run RMAN at a time when it will interfere less with the
application.
Consider running a 10046 trace on the application with and without RMAN
executing, and see what wait times are different. This may help you pinpoint
the interference.
Also, gather statistics at the O/S level with and without RMAN running. I
would expect many of the most important effects might not be seen within
Oracle itself, but only at the O/S.
Dennis Williams
DBA
Lifetouch, Inc.
-----Original Message-----
From: oracle-l-bounce@xxxxxxxxxxxxx [mailto:oracle-l-bounce@xxxxxxxxxxxxx]
On Behalf Of Fuad Arshad
Sent: Monday, December 20, 2004 8:37 AM
To: oracle-l@xxxxxxxxxxxxx
Subject: Rman io done waits and the rate parameter
we are having very high iodone waits.
and i was wonderign if somone on a list has had to go thru the same issues
thisis a snap from the statspack reports.
Avg
Total Wait wait
Waits
Event Waits Timeouts Time (s) (ms)
/txn
---------------------------- ------------ ---------- ---------- ------
--------
io done 7,318,903 5,192,927 122,807 17
146.5
sbtwrite2 4,240,771 0 126,120 30
84.9
io done 7,318,903 5,192,927 122,807 17
146.5
currently we are on gen1 tape drive plannign to move to gen2's
the backup of a 1 tbte database takes around 4 hours but the application
timings go up by 20%
i am also looking at ways people have had to use the rate or the readrate
parameter to throttle down the amount of i/o and tune down rman.
--
//www.freelists.org/webpage/oracle-l
--
//www.freelists.org/webpage/oracle-l
Other related posts:
