Re: Backup using only one CPU - SA wants to know why
Date: Tue, 13 Nov 2012 08:26:58 -0700
Message-ID: <CAJzM94CyJhbz=1w5Q6GSWaO6SSnP+qzTjz-+78Y0=Ztp8-=Zzw_at_mail.gmail.com>
My backups are to disk, not tape, so that isn't an issue. What I'm seeing is lots of I/O, as expected, and very little CPU, also as I would expect. I personally do not consider this an issues since the backups are performing well and finishing several hours before the end of the backup window. The SA has some monitoring tool he uses on the IBM mainframe. Occasionally, he uses the top command.
Thanks for the suggestions.
Sandy
On Tue, Nov 13, 2012 at 4:23 AM, Martin Berger <martin.a.berger_at_gmail.com>wrote:
> Sandy,
>
> do if you have proper licenses, can you check v$active_session_history /
> dba_hist_active_sess_history what the rman sessions are doing during
> backup?
>
> That should give you a reasonable estimation where they spend their time.
> (probably better data than a unspecific claim.)
>
> hth
> Martin
>
>
>
>
>
> On Mon, Nov 12, 2012 at 9:56 PM, Sandra Becker <sbecker6925_at_gmail.com>wrote:
>
>> I have a hunch it's something along those lines, but don't know how to
>> prove it. My SA insists there's a setting in the database to "turn on
>> multiple CPUs". If it already recognizes there are 5 of them, I'm not
>> sure
>> what else I can do.
>> Sandy
>>
>> On Mon, Nov 12, 2012 at 1:53 PM, <Christopher.Taylor2_at_parallon.net>
>> wrote:
>>
>> > According to my AIX admin (on a similar question I asked him about
>> active
>> > sessions) he said its because of the way IBM does CPU scheduling.
>> >
>> > I wish I had more info on that.
>> >
>> > Chris
>> >
>> > -----Original Message-----
>> > From: oracle-l-bounce_at_freelists.org [mailto:
>> oracle-l-bounce_at_freelists.org]
>> > On Behalf Of Sandra Becker
>> > Sent: Monday, November 12, 2012 1:39 PM
>> > To: oracle-l
>> > Subject: Backup using only one CPU - SA wants to know why
>> >
>> > Oracle 11.2.0.2
>> > SLES11 on IBM mainframe
>> > DB size - just below 1T
>> > I run an incremental 0 backup on my production OLTP once a week,
>> > incrementals the rest of the week. Recently changed the disk_asynch_io
>> to
>> > TRUE and the filesystemio_options to SETALL, per recommendations from
>> > Oracle. I then added a channel to the backup scripts so it uses 4
>> > channels. These changes shaved 35 pecent off the time it takes to run
>> the
>> > incr0 backup, which I am very happy with.
>> >
>> > The SA monitoring the performance of the mainframe says the incremental
>> 0
>> > backup of our production OLTP is pegging a single CPU, out of 5
>> available.
>> > He thinks the backup would run even faster if it used more CPUs and
>> > suggested setting the PARALLELISM parameter to more than 1.
>> Unbeknownst to
>> > him, it had been set to 3 for the past 6 years. He wants to know why
>> the
>> > backup isn't using more CPUs. Any ideas/suggestions?
>> >
>> > --
>> > Sandy
>>
>
-- Sandy Transzap, Inc. -- http://www.freelists.org/webpage/oracle-lReceived on Tue Nov 13 2012 - 16:26:58 CET