Oracle FAQ | Your Portal to the Oracle Knowledge Grid |
Home -> Community -> Usenet -> c.d.o.server -> Re: Changing to Raw Redo
"RSH" <RSH_Oracle_at_worldnet.att.net> wrote in message news:<Wagf8.4412$106.252464_at_bgtnsc05-news.ops.worldnet.att.net>...
> Uh, remapping devices and things would be a problem; but more to the point,
> why do redo in raw partitions? Redos are written to serially, till they're
> filled, then a log switch, etc. Have you done some analysis that would lead
> you to conclude that putting your redo log files on raw partitions would
> alleviate a particular problem?
The decision was based on a series of tests. There are a number of in-depth documents recommending raw for redo. You can check Steve Adam's site (ixora.com.au). He has a very detailed document on this. I also have a copy of Oracle's internal performance tuning literature which suggests the same.
My own tests showed 50% of deduction on "redo write time" after switching to raw. I performed 5 tests before and 5 tests after and the numbers are very consistently convincing. Of course the tests are a bit unrealistic, and it does show you the magnified difference. In real life the difference will definitely be smaller, but the point is, there IS a difference attractive enough to go thru the inconvenience of raw. At least I am convinced.
>
> As far as propagation by (I assume you mean) Oracle Standby, I'm not sure
> whether it supports raw partitions but I'd be surprised if it did not; you
> never know, though...
>
I know Standby supports raw redo. It has no difference than primary. In fact, while a standby is recovering, it does NOT have any redo logs. As it deos not generate redo while recovering. But when you open standby, it has to create redo logs at that point. And it is at that point you will see failure, because the raw paths do not exist on the standby box. So my question is, can I use a symbolic link to point to the raw chunk in the first place, so when I open the standby, I can point the link to another raw chunk on the standby box ?
The other way around, I believe, is to create controlfile,( instead of using standby controlfile ) and modify the redo paths in there, and open resetlogs. Does anybody know if this works ?
Trying to drop the logfile and add new one does not work, because database is not open and you can not open the database unless Oracle can create the redo logs as specified in the standby controlfiles.
> and what is your platform?
8.1.7 on Solaris 2.7 Received on Thu Feb 28 2002 - 16:44:25 CST