Re: async i/o via nfs?

"Billy Verreynne" <vslabs_at_onwe.co.za> schrieb im Newsbeitrag news:1a75df45.0308052232.6ff59c0f_at_posting.google.com...
> "Volker Hetzer" <volker.hetzer_at_ieee.org> wrote
>
> > > Shudder. How on earth did you decide to locate one of your controlfile
> > > copies on a nfs-mounted drive? Why not all of them?
>
> > I've got three copies, one near the db files, one on another volume and
one
> > on another machine. So far it has worked perfectly.
>
> I concur with Sybrand. It is IMO a very bad idea to have a control
> file on a NFS mount.

All right, all right, I haven't got any bad advice here yet, I'm seriously thinking about it.

>
> > What would you suggest for having a copy on another machine?
>
> Local copy. Then pick from rdist, remote copy, ftp, etc. There are
> numerous utilities that can make that easy and painless.
Ok, but in order to guarantee consistency wouldn't it be better to back it up first using ALTER DATABASE BACKUP CONTROLFILE TO ...; and rdist then?
And still that always leaves me with an out of date controlfile.

> > I've just had a look and what I got was "RAC on Linux Best Practices"
> > and it says Asynchronous I/O can be used on whatever and nfs too.
>
> Async i/o for control file i/o on NFS is IMO meaningless. It adds no
> technical or performance benefits.

Of course it doesn't. What I was interested in knowing was wheather the db in async mode *tolerates* the nfs for controlfiles. After all, I can't say "async=true-except-for-control-files", so if oracle used async i/o for controlfiles (and perhaps archiving logs too) and crashed when the target is on a nfs mount I couldn't use async i/o for datafiles either.

Lots of Greetings and Thanks!
Volker Received on Wed Aug 06 2003 - 11:22:39 CDT