Re: Oracle on a NetApp

In article <32EE2438.4D62_at_pencom.com>, Michael Karafotis <karafoti_at_pencom.com> writes
>Okay everyone,
>
>Could someone explain to me why it is bad to store oracle datafiles,
>redologs, and control files on a Network Appliance?
>
>As I understand it the argument against doing such a thing is
>consistency of the redo logs. Oracle needs to be sure that data is
>actually written to the device before it can continue on and clear the
>redo buffer. My understanding of NFS V3 and V2 for that matter is that
>all writes are by default synchronized (in other words dirty writes or
>cache writes do not return a write success, only a stable write to the
>device does.)

I take this to mean that you will know when a write fails. This is no good at all. An Oracle instance will crash if it gets a write failure. I suspect that the incidence of network failures is much higher than the incidence of disk failures.

> If this is the case why is it taboo to run the database
>off a FAST, RAID 5, backend (private subnet) NFS server like a Network
>Appliance.
>

The other big issue is performance. Networks are much slower than IO busses.

People do run non-critical (read trivial) databases on NFS, but anyone who commits anything important to such a system needs their head examined.

-- 
Jim Smith