Re: 10.2 NFS mount points
- From: Mladen Gogala <gogala@xxxxxxxxxxxxx>
- To: marquezemail@xxxxxxxxx
- Date: Wed, 01 Feb 2006 22:38:56 -0500
On 02/01/2006 09:35:25 PM, Chris Marquez wrote:
> All this talk about Oracle & NFS. It has been a long time since I looked,
> but hasn't Oracle always said "NO" to NFS (datafiles)? I'm not talking
> about NAS like NetApp...I'm talking about setting up a home grown NFS box
> and putting datafiles on it?...shoot I'm not even sure they support many
> CFS's?
Chris, the words of wisdom about datafiles and NFS were spoken a long
time ago, ding the era of NFS v2. Since then, somewhere in 1994, SUN
published specification for NFS v3. Differences were many, but among
the two most important were locking and the ability to use TCP connection.
Previously, in v2 protocol, NFS was using UDP and was stateless, just like
http is today. When something is stateless, it is rather hard to lock it,
so NFS v2 did not support locking. That would make it rather hard to
operate a database. In addition to locking, NFS4 will add client side
caching, which would speed things up enormously and async I/O and
direct I/O will come standard.
Some of it was previously available on Solaris (see cfsadmin) but not
on most of the other unix versions as this features needed changes in
the kernel as kernel is the part which performs I/O and decides about
caching data blocks. So now with NFS4 you have async and direct I/O
and client caching. Not much difference from a local disk at all.
The old lore no longer applies. Really good summary can be found at:
http://www.citi.umich.edu/projects/nfsv4
Please, feel enlightened, by all means.
--
Mladen Gogala
http://www.mgogala.com
--
//www.freelists.org/webpage/oracle-l
Other related posts:
