RE: Is nfs reliable?

Ok Let me clarify if you setup and manage nfs mounts correctly they are extremely stable. If you do not setup the nfs mounts correctly in my experience they are not reliable. If you are going to use NFS mounts to support your environment please spend the time and set them up correctly.  

Thanks, Matt  

From: oracle-l-bounce_at_freelists.org [mailto:oracle-l-bounce_at_freelists.org] On Behalf Of Mark W. Farnham Sent: Friday, June 3, 2016 5:20 PM
To: tim_at_evdbt.com; mvshelton_at_chartermi.net; rfreeman_at_businessolver.com; 'Michael Cunningham' Cc: 'oracle-l_at_freelists org'
Subject: RE: Is nfs reliable?  

First, I agree with and applaud Tim regarding his careful and accurate description of NFS, TCP, and UDP.  

Second, I’ll note that Matt added the word “mounts” to his comment. IF it is the mounts specifically to which Matt refers, there has been a spotty history over time and releases of various operating systems about whether you could rely on particular mount point naming, addressing, and characteristics across normal and failure mode operating system reboots. I believe that is mostly in the past, but I cannot assure you Matt is wrong, because a) no one can enumerate that they are reliably present across all combinations, and b) Matt could be referencing particular IT shop mount point activation processes that were apparently broken at a particular organization.  

But back to Tim’s point: If you chose the right protocol (and mount reliably) you will be up to industry standard on the reliability front.  

As Tim mentioned earlier in the thread it is quite possible to configure nfs over transport that is “lossy.” If you are using Oracle that would be a poor choice for at least database files and redo logs. Likewise, a bone head order of reboot could start oracle before some of the required mount are made and in reboot software that forks parallel operations in the restart hilarity can result if restarting Oracle is not definitely serially after establishing the mount points on which it relies and you thereby produce a race condition that appears to the casual observer to not be deterministic. So nfs CAN be unreliable, but it does not have to be and it is not rocket science to get it right. Likewise nfs mounts.  

Sorry to be long-winded, but I really didn’t want any folks who have observed “nfs mount point failure” and therefore “know” nfs is not reliable to become confused as to whether Tim was wrong in any regard about his careful and accurate description of NFS, TCP, and UDP.  

mwf  

From: oracle-l-bounce_at_freelists.org [mailto:oracle-l-bounce_at_freelists.org] On Behalf Of Tim Gorman Sent: Friday, June 03, 2016 4:37 PM
To: mvshelton_at_chartermi.net; rfreeman_at_businessolver.com; 'Michael Cunningham' Cc: 'oracle-l_at_freelists org'
Subject: Re: Is nfs reliable?  

Matt,

I respectfully disagree; there is indeed an easy answer.

First, the term "reliable" is ambiguous and must be clarified before engaging in a meaningful discussion. In this discussion, I believe the term "reliable" is not defined as the "absence of failure", but instead as "loss-less transmission of data", with "loss-less" meaning "complete and accurate".

TCP and UDP are alternative protocols residing at layer 4 (transport) of the 7-layer OSI model <https://en.wikipedia.org/wiki/OSI_model> . The first three layers of the OSI model define the hardware carrier, structure, framing, addressing, and routing of data, and it is at the transport layer that the branch between loss-less and loss-y session protocols is implemented.

NFS is implemented at level 7 (application) of the model, and so the question of "reliability" (as I believe we are discussing it) has been resolved several layers down.

In any context (analog, digital, or human), "reliable" never means "perfect". Error, loss, and corruption is always possible for a myriad of reasons, applying equally to any protocol. TCP is "loss-less" by design, but no design is absolutely perfect.

Hope this helps!

-Tim

On 6/3/16 13:17, Matt wrote:

I do not think there is an easy answer whether or not nfs mounts are reliable. IT depends upon your organization infrastructe teams. I have worked at organizations were my nfs mounts were completely unreliable. My current organization I cannot remember the last time I have had an issue with my nfs mounts.  

Thanks, Matt  

From: oracle-l-bounce_at_freelists.org [mailto:oracle-l-bounce_at_freelists.org] On Behalf Of Robert Freeman Sent: Friday, June 3, 2016 1:31 PM
To: Michael Cunningham
Cc: oracle-l_at_freelists org
Subject: RE: Is nfs reliable?  

I would not have an issue with monitoring scripts… However, anything diagnostic or the like I’d keep on local storage…  

From: Michael Cunningham [mailto:napacunningham_at_gmail.com] Sent: Friday, June 03, 2016 11:30 AM
To: Robert Freeman <rfreeman_at_businessolver.com> Cc: oracle-l_at_freelists org <oracle-l_at_freelists.org> Subject: Re: Is nfs reliable?  

Robert, how would you feel about having a library of monitoring scripts on an nfs mount? I do this so that scripts are available on every Oracle server. I also backup config files, etc. to this mount making it easy for me to scan, in a single directory, something like all cron backup files to see where something is running. It's these things why I feel the nfs mount serves a great benefit. And many others...  

Michael  

On Fri, Jun 3, 2016 at 8:34 AM, Robert Freeman <rfreeman_at_businessolver.com> wrote:

Mike,  

We Run our RAC databases using NFS as the shared media for datafiles. We have yet to have a failure because of NFS. We don’t use shared software directories however, just local directories.

My personal preference is for local software.

RF    

From: oracle-l-bounce_at_freelists.org [mailto:oracle-l-bounce_at_freelists.org] On Behalf Of Michael Cunningham Sent: Friday, June 03, 2016 12:02 AM
To: oracle-l_at_freelists org <oracle-l_at_freelists.org> Subject: Is nfs reliable?  

I had someone tell me today that nfs should not be relied on and it should not be used for a shared mount that needed to be reliably available.  

Has anyone ever hear this before?  

--

Michael Cunningham  

--

Michael Cunningham  

--

http://www.freelists.org/webpage/oracle-l Received on Sat Jun 04 2016 - 01:08:35 CEST