I am not a system admin, so I am not very familiar with NFS, but I was
told that it has various problems (many enough so Oracle Corp. refuses
to support you when they find out that you are running Oracle database
with NFS).

Is there some other safe/stable/secure/fault-tolerant alternative to
NFS available in Linux? I need to set up some way to have multiple
machines share the same file systems. Thank you in advance.

-----BEGIN PGP SIGNED MESSAGE-----
Hash: SHA1

On 2004-02-27, Forte Agent <1443131@usenetplanet.com> wrote:
In general, no heavy RDBMS should be run over NFS. That doesn't mean
NFS doesn't have valid applications; just don't share your database
over NFS.

If you're on a relatively closed and safe network, just use NFS.
There's no alternatives that are relatively well-established yet.
Some look promising (I've heard some good things about Coda), but aren't
really in wide use yet.

- --keith

- --
kkeller-usenet@wombat.san-francisco.ca.us
(try just my userid to email me)
AOLSFAQ=http://wombat.san-francisco.ca.us/cgi-bin/fom

-----BEGIN PGP SIGNATURE-----
Version: GnuPG v1.2.3 (GNU/Linux)

iD8DBQFAP7vghVcNCxZ5ID8RAkpjAJ9ZkjrPDs29smb93yK2pY CmvX/NBQCgjajJ
P/iVsdT8NREcnJla8e42aHY=
=MAzL
-----END PGP SIGNATURE-----

Keith Keller wrote:
It also depends on how you're running ORACLE. If the ORACLE
processes are running on the same machine as the NFS server
then Oracle shouldn't care because all ORACLE disk accesses are
local to the machine Oracel is running on.

If the Oracle procs are running
on one machine and the DBMS is on the NFS server, you're
probably shooting yourself in the foot, performance-wise.
(And could very well be a part of the reason Oracle
won't support it.)

From what I have heard second and third hand, tho,
NFS is neither fault-tolerant nor particularly stable.
Hangs have been reported which require server restart.
But, as I said, these are war stories which may or
may not be exagerating things and may depend on your
definition of fault-tolerant. Mine is <1 hour per
year down-time. YMMV.

I would re-consider your partitioning of the functionality
between your machines, if this is an option.



Ñ
"It is impossible to make anything foolproof because fools are so
ingenious" - A. Bloch

Quoth Forte Agent <1443131@usenetplanet.com>:
Are you planning to run a database on top of it?

Don't. Databases _really_ should be run on local disk.
--
output = ("cbbrowne" "@" "cbbrowne.com")
http://cbbrowne.com/info/nfs.html
"The classic Common Lisp defmacro is like a cook's knife; an elegant
idea which seems dangerous, but which experts use with confidence."
-- Paul Graham, _On Lisp_

Centuries ago, Nostradamus foresaw when Keith Keller <kkeller-usenet@wombat.san-francisco.ca.us> would write:
I wouldn't expect Oracle to be one iota happier about supporting
someone running their database atop Coda or any other distributed
filesystem.

You should only be running databases atop NFS if you're using some
form of SAN which has the side-effect that you'll be doing "rocket
science" stuff that mandates having a _really_ competent sysadmin
around.

NFS is a perfectly reasonable choice for handling the need to share
files across servers. Just don't take it too far...
--
If this was helpful, <http://svcs.affero.net/rm.php?r=cbbrowne> rate me
http://cbbrowne.com/info/wp.html
MICROS~1: The company that brought new meaning to "Nervous System"

Nick Landsberg <hukolau@NOSPAM.att.net>,
In a message on Fri, 27 Feb 2004 22:50:22 GMT, wrote :

NL> Keith Keller wrote:
NL> >=20
NL> > On 2004-02-27, Forte Agent <1443131@usenetplanet.com> wrote:
NL> >=20
NL> >>I am not a system admin, so I am not very familiar with NFS, but I was
NL> >>told that it has various problems (many enough so Oracle Corp. refuses
NL> >>to support you when they find out that you are running Oracle database
NL> >>with NFS).
NL> >=20
NL> >=20
NL> > In general, no heavy RDBMS should be run over NFS. That doesn't mean
NL> > NFS doesn't have valid applications; just don't share your database
NL> > over NFS.
NL> >=20
NL> >=20
NL> >>Is there some other safe/stable/secure/fault-tolerant alternative to
NL> >>NFS available in Linux? I need to set up some way to have multiple
NL> >>machines share the same file systems. Thank you in advance.
NL> >=20
NL> >=20
NL> > If you're on a relatively closed and safe network, just use NFS.
NL> > There's no alternatives that are relatively well-established yet.
NL> > Some look promising (I've heard some good things about Coda), but aren'=
NL> t
NL> > really in wide use yet.
NL> >=20
NL> > - --keith
NL>
NL> It also depends on how you're running ORACLE. If the ORACLE
NL> processes are running on the same machine as the NFS server
NL> then Oracle shouldn't care because all ORACLE disk accesses are
NL> local to the machine Oracel is running on.
NL>
NL> If the Oracle procs are running
NL> on one machine and the DBMS is on the NFS server, you're
NL> probably shooting yourself in the foot, performance-wise.
NL> (And could very well be a part of the reason Oracle
NL> won't support it.)
NL>
NL> From what I have heard second and third hand, tho,
NL> NFS is neither fault-tolerant nor particularly stable.
NL> Hangs have been reported which require server restart.
NL> But, as I said, these are war stories which may or
NL> may not be exagerating things and may depend on your
NL> definition of fault-tolerant. Mine is <1 hour per
NL> year down-time. YMMV.

Most of these 'war stories' are probably about things in the 'olden
daze'. Generally modern NFS software is quite stable. I am not sure if
'fault-tolerant' particularly applies.

The main source of issues with NFS is the fault-tolerance of the NFS
server itself. That and 'cross-mounting': eg machine A nfs serves file
system /A/foo to machine B and machine B nfs serves file system /B/bar
to machine A, you get a chicken and egg problem -- both machines are
likely to hang trying to mount each other's disks, unless one is real
careful (the 'bg' option is a must here). If one machine fails or
crashes or is shutdown/rebooted, the other *might* hang.

If you have one (or a small number of) NFS server(s) and have this
(these) machine(s) on a UPS, with hot swappable disks, dual redundant
power supplies, RAID-5, etc., you should not have stability or fault
issues, so long as you are careful and sensible (don't randomly unmount
file systems on the server, etc.).

NL>
NL> I would re-consider your partitioning of the functionality
NL> between your machines, if this is an option.
NL>
NL>
NL>
NL> =D1
NL> "It is impossible to make anything foolproof because fools are so=20
NL> ingenious" - A. Bloch
NL>
NL>

\/
Robert Heller ||InterNet: heller@cs.umass.edu
http://vis-www.cs.umass.edu/~heller || heller@deepsoft.com
http://www.deepsoft.com /\FidoNet: 1:321/153

Christopher Browne <cbbrowne@acm.org>,
In a message on 28 Feb 2004 00:19:14 GMT, wrote :

CB> Quoth Forte Agent <1443131@usenetplanet.com>:
CB> > I am not a system admin, so I am not very familiar with NFS, but I was
CB> > told that it has various problems (many enough so Oracle Corp. refuses
CB> > to support you when they find out that you are running Oracle database
CB> > with NFS).
CB> >
CB> > Is there some other safe/stable/secure/fault-tolerant alternative to
CB> > NFS available in Linux? I need to set up some way to have multiple
CB> > machines share the same file systems. Thank you in advance.
CB>
CB> Are you planning to run a database on top of it?
CB>
CB> Don't. Databases _really_ should be run on local disk.

Light-duty PostgreSQL databases seem to run just fine over NFS. Visit
http://vis-www.cs.umass.edu/ and select 'Publications'. You are now
talking to database driven pages. The database server (and web server)
are running on a 350mhz Celeron. The database file system is NFS
served by a dual 2.4Gig P4htZeon file server. These two machines have
their own 'private' 100BaseT network.

CB> --
CB> output = ("cbbrowne" "@" "cbbrowne.com")
CB> http://cbbrowne.com/info/nfs.html
CB> "The classic Common Lisp defmacro is like a cook's knife; an elegant
CB> idea which seems dangerous, but which experts use with confidence."
CB> -- Paul Graham, _On Lisp_
CB>

\/
Robert Heller ||InterNet: heller@cs.umass.edu
http://vis-www.cs.umass.edu/~heller || heller@deepsoft.com
http://www.deepsoft.com /\FidoNet: 1:321/153

The world rejoiced as Robert Heller <heller@deepsoft.com> wrote:
That databases can "seem to run just fine" is not the issue. Fault
tolerance and crash recovery are the issues.

I normally go with Bruce Momijan's recommendations on this:

<http://candle.pha.pa.us/main/writings/pgsql/hw_performance/node13.html>

"NFS and other remote file systems are not recommended for use by
POSTGRESQL. NFS does not have the same file system semantics as a
local file system, and these inconsistencies can cause data
reliability or crash recovery problems."

That's more than likely the reason why Oracle told the O.P. that they
would refuse to support his system. For practically any database
system, the "crash recovery" capabilities will be corrupted to at
least some degree by the differences in semantics between NFS and
local filesystems.
--
If this was helpful, <http://svcs.affero.net/rm.php?r=cbbrowne> rate me
http://www.ntlug.org/~cbbrowne/fs.html
Rules of the Evil Overlord #18. "I will not have a son. Although his
laughably under-planned attempt to usurp power would easily fail, it
would provide a fatal distraction at a crucial point in time."
<http://www.eviloverlord.com/>