Re: Documentation on Transaction Management
From: Mathias Magnusson <mathias.magnusson_at_gmail.com>
Date: Fri, 20 Nov 2009 06:32:30 +0100
Message-ID: <8580d4110911192132w5e196232pdd1b4a6078a69ba_at_mail.gmail.com>
How long time does it take to run the update as a single update? Sure enough undo would need to be allocated, but this would show if the problem is with the approach or with the database (number of indexes and such). It could also show that the simple way actually works fast enough to do the job.
Date: Fri, 20 Nov 2009 06:32:30 +0100
Message-ID: <8580d4110911192132w5e196232pdd1b4a6078a69ba_at_mail.gmail.com>
How long time does it take to run the update as a single update? Sure enough undo would need to be allocated, but this would show if the problem is with the approach or with the database (number of indexes and such). It could also show that the simple way actually works fast enough to do the job.
If that turns out to be too slow, then one possible solution would be to have multiple parallel jobs operate on different parts of the table.
Can new rows be added to the table while the update runs?
Mathias
On Fri, Nov 20, 2009 at 3:16 AM, rjamya <rjamya_at_gmail.com> wrote:
> 10m rows should not be more than a few minutes unless there are lots of
> active transactions and/or indexes. I'd try to minimize outside lookups and
> by scanning the table (to be updated) first, get values from lookup tables
> and cache them locally then run simple updates.
>
> Then again we are talking about lack of more info :)
>
> Raj
>
>
>
-- http://www.freelists.org/webpage/oracle-lReceived on Thu Nov 19 2009 - 23:32:30 CST