Re: Frequent log switches during sqlldr batch jobs

From: Frank van Bortel <fvanbortel_at_netscape.net>
Date: Wed, 02 Feb 2005 22:39:41 +0100
Message-ID: <ctrh7t$gro$1@news5.zwoll1.ov.home.nl>

Thorsten Jens wrote:
> Frank van Bortel wrote:
>

>>thojens_at_gmx.de wrote:
>>
>>>Hi,
>>>
>>>we are running 9.2.0.6 on Solaris. Each night, a bunch of sqlldr

>
> batch
>

>>>jobs are run and the redo log switches per hour increase

>
> dramatically:
>

>>>100 and more per hour, while during normal production hours we're

>
> at a
>

>>>maximum of about 25 (usually much lower).
>>
>>The amount of data written would indeed be the same; there is some

>
> O/S
>

>>overhead in closing a file, and opening a new one (new inode entry),
>>but I wouldn't worry about it, unless you see switches within a

>
> second.
>
> OK.
>
>

>>Query optimal_logfile_size in V$instance_recovery.

>
> Not available in 9i, I'm afraid.
>
> Thorsten
>

Take about 5 switches per hour, then. Gives you 12 minutes worth of transactions to loose when the current redolog file crashes.
Large redo log files make instance recovery slower (more transactions need to be replayed).

There is a part about redo log tuning in the 9i tuning manuals.

-- 
Regards,
Frank van Bortel

Received on Wed Feb 02 2005 - 15:39:41 CST