Oracle FAQ Your Portal to the Oracle Knowledge Grid
HOME | ASK QUESTION | ADD INFO | SEARCH | E-MAIL US
 

Home -> Community -> Usenet -> c.d.o.server -> Re: Scalable Performace - Inserts/Updates

Re: Scalable Performace - Inserts/Updates

From: Nuno Souto <nsouto_at_nsw.bigpond.net.au.nospam>
Date: Thu, 21 Dec 2000 09:15:38 GMT
Message-ID: <3a41c59c.1160812@news-server>

On Wed, 20 Dec 2000 17:27:07 GMT, Allan Plesniarski <aplesnia_at_my-deja.com> wrote:

>My understanding is that Oracle can corrupt one member of a redo log
>group, while the other member is written to cleanly.

Hmmm yes, possible. I've seen bugs before that corrupted ALL members, particularly when handling LONG type columns. Earlier versions. My contention here is that using more than one member when the hardware is already mirroring does not really buy you any additional protection. And it buys you a hit on performance.

As for the best protection, there is nothing better than enabling the log checksum thing. But watch the performance go out the window when you do that! Actually, there may be a way around that problem if you're using EMC boxes, but that's out of scope here.

> Why one and not
>the other, I do not know. However, I did get hit with redo log
>corruption in the past. And guess what, I believed in only hardware
>mirroring, so there was only one redo log member per group.

I'll bet you had long columns somewhere in there! Was it AF? Or Oracle Financials?

>Embarrassingly, I was reminded at the time by a colleague that had the
>redo log groups had more that one member, chances were that only one
>redo log member would be corrupted and the database could continue
>normal operation.

If your corruption was caused at redo log generation as I suspect, it wouldn't have helped you one bit. But it's always very hard to determine where the cause was, anyway.

>Operating a database without more than one redo log member per group is
>ok as long as you understand and accept the risks.
>

Of course, you're quite right. When pushing for 1000 commits/sec, you gotta cut corners somewhere. Or else spend BIG bucks on really BIG iron. I believe that an EMC or equivalent will give us a more than adequate level of protection that doesn't require the multiplexing of the redo logs. Others might find that level of protection unacceptable. Hey, that's why you can use it or not ain't it? ;-)  

Cheers
Nuno Souto
nsouto_at_bigpond.net.au.nospam
http://www.users.bigpond.net.au/the_Den/index.html Received on Thu Dec 21 2000 - 03:15:38 CST

Original text of this message

HOME | ASK QUESTION | ADD INFO | SEARCH | E-MAIL US