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: Rman - Curious Question

Re: Rman - Curious Question

From: Joel Garry <joel-garry_at_home.com>
Date: 28 Jul 2005 14:48:43 -0700
Message-ID: <1122587323.734786.182800@g49g2000cwa.googlegroups.com>

Holger Baer wrote:
> amerar_at_iwc.net wrote:
> > Hi All,
> >
> > Just a curious question:
> >
> > When you are duplicating a database which is on a differnt host, why
> > does it require you to copy the files to the remote host? Why can't
> > RMAN use the files on the primiary host?
> >
> > If you are using a recovery catalog, you use the catalog on the primary
> > host, so, why not also use the backup pieces? All the manual says is
> > that it requires you to copy the file.....
> >
>
> There is (at least) one misconception in your post: The catalog does only hold
> metainformation about the backups, so the fact that it is accessible
> is rather irrelevant for the question why the backup pieces ain't used.
> (In fact, it is recommended that the catalog and the target database are
> on different HW, so the files wouldn't be accessible through the catalog db).
>
> So the backup files are accessible primarily through the target
> - except if they are on tape and both machines can access the tape library
> - except if they are on a filesystem accessible to both machines.
>
> Why does Oracle not use Net 8? Because RMAN creates a server process on the
> target db and no-one in their right minds would like the production server
> to flood his network interface with the 1 TB you mention later in this
> thread.

This part I don't get. What is the difference between flooding the interface with Net8, flooding it with ftp or flooding it with nfs? It's still a T of data plus overhead, and you don't want to fight the phone system over bandwidth for it.

I find the NFS method Sybrand quoted odd, as I think any NFS data connection is suspect unless the hardware and software is specifically OK'd by Oracle, so a flat "you can do it this way" would be wrong.

I'm sure at least part of the disagreement among the docs has to do with the range of NFS implementations. Certainly any that are UDP based wouldn't reliably work (guess what the U stands for - OK, it stands for User, not Unreliable, although http://www.faqs.org/rfcs/rfc768.html says "delivery and duplicate protection are not guaranteed" ). And some modern NFS still are UDP (for example, hp-ux 11i by default trys TCP first, then falls back to UDP on failure). I've even seen rcp have problems with files over, say, 20G (on a fast local network).

>
> I hope this clears the smoke

The fan seems to be making more smoke :-)

jg

--
@home.com is bogus.
http://www.signonsandiego.com/uniontrib/20050727/news_m1m27tfdmar.html
Received on Thu Jul 28 2005 - 16:48:43 CDT

Original text of this message

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