Re: Frequent log switches during sqlldr batch jobs

From: Frank van Bortel <fvanbortel_at_netscape.net>
Date: Tue, 08 Feb 2005 20:32:51 +0100
Message-ID: <cub41q$4j1$1@news6.zwoll1.ov.home.nl>

xhoster_at_gmail.com wrote:
> "Thorsten Jens" <thojens_at_gmx.de> wrote:
>

>>Frank van Bortel wrote:
>>
>>>Take about 5 switches per hour, then. Gives you 12 minutes
>>>worth of transactions to loose when the current redolog file
>>>crashes.
>>
>>If I sized the redo logs to have ~5 switches per hour during high load
>>times, then I'd have almost no switches at all during the day -- which
>>would mean a lot of lost work in the worst case.

>
>
> I've seen this argument many times, but I don't understand it. Isn't
> the worst case the one where both the online redo log and recently archived
> redo logs are lost? In most systems I see, the online redo logs are
> protected by mirroring and multiplexing to the same extent as the freshly
> archived rego logs are (if not more). Is this not usually the case?
>
> Xho (not a backup/recovery DBA)
>

Both? What do you mean both online redo logs? I have at least 3 groups (test/dev: 3 files, not groups). Not that I set standards, but it is regarded Common Sense(tm) to have at least three (one being written, one being archived, one ready to be overwritten).

And of course, the more you lose, the worse it is. Guess the worse case would be a fire, destroying your hardware: all online and not backed up archived redo logs get lost (and hopefully your tapes are off site!!). You loose all transactions from the point where your last backups of the archived redo log files is.

Sizing also comes in place, where unexpectedly the archive log destination gets filled up: you will only have your online redo log groups to fill, before you come to a halt - I've seen systems with about 12 hours worth of online redo log groups.

-- 
Regards,
Frank van Bortel

Received on Tue Feb 08 2005 - 13:32:51 CST