Re: raw devices

Hi Jerry,
Thanks for answering. What I was refering to was some sort of raid 10 that you can construct on Solaris at the software level. And from the documentation we inferred that this is safer than the file system. And I still think it's true(you didn't contradict me either).It's also true that you can get a false sense of security by buying expensive raid and ups and placing your server in a room with lots of sprinklers. The theory that I read was that for the file system the OS will report completion for a write operation after writing it in cache, for a raw device after writing it to disk.
Regards,
Max Mera

Jerry Gitomer wrote:

> Hi,
>
> You may be suffering from a false sense of security when it
> comes to RAW devices. It doesn't matter how smart the RDBMS is
> the OS can take control away from the RDBMS without the RDBMS
> ever knowing about it. In fact when using an LVM or hardware
> RAID there is a good chance that all of your I/O is buffered --
> including writes to raw devices.
>
> In either case the answer is a UPS and RAID using mirroring
> or parity. The former will protect you from a power failure and
> the latter will protect you against a disk crash. Finally, in
> order to eliminate the probability of getting nailed by a server
> failure go for a system that can deliver 99.999% uptime.
>
> regards
> Jerry Gitomer
>
> DM wrote in message <37B977FB.F2517CE_at_auto.com>...
> >Hello,
> >I know that this was part of a previous thread, but the thread
> didn;t
> >cover the database integrity.
> >Somebody asked if you should use cooked or raw devices with
> Oracle, for
> >speed and integrity reasons.
> >From what I know, If you use cooked devices on unix, the file
> system
> >will try to cache as much as it can and the data you commit will
> be
> >"commited" in the fs cache and flushed from time to time to
> disk. This
> >means that if the server crashes you will probably loose
> non-flushed
> >data with the database engine not being aware of the loss(so no
> rollback
> >for you).
> >Am I wrong, is Oracle smart enough to detect this situation?
> >Thanks
> >Max Mera
> >
Received on Tue Aug 17 1999 - 13:47:21 CDT