Oracle database over NFS !!

[KapilRaj]

Guys,

I don't like this configuration. I have had my bad days working on the NFS with application filesystems. Now here I am , where I am going to put oracle datafiles on NFS mounted filesystems :(

The NFS Server is a NAS (EMC Celerra). The oracle database could be as big as 40-50 GB.

Do you think I should strongly disagree to this configuration ?.

I thought the write speeds are going to be bad, but when I did a test now it gives a speed close to directly attached SCSI storage. Ok, the test filesystem is the only filesystem present on the NAS now.

Give it a thought, and let me know what would do if you were in my shoes ?

Regds,

Kaps

[Steven E. Protter · Accepted solution]

Shalom Kaps,

I'm so against this configuration.

However, we tested this configuration and it worked just fine.

If you db is write intensive you might want to be on the same fiber network.

Bottom line is we battle tested it, it worked and Oracle supports the setup.

SEP

[Phillip Thayer]

Wow, I would really be against a configuration like this.

What happens when the NFS Server on the EMC system fails in the middle of an Oracle transaction commit? That ould probably corrupt the DB and cost time and money to the business to recover the DB.

The question I would pose is what is the cost of getting the system connected diredctly to the EMC SAN now as opposed to the cost of possibly recovering the DB from a backup tape and applying all the transactions since the backup tape was made?

The one thing that managment understands is money, what t hey fail to understand is when they will have to spend it. Spend less now to do it right or spend more later to keep the business running so they can spend the money to it the right way and migrate data and applications. Guess which one costs more money?

Phil

[A. Clay Stephenson]

While it is certainly not my first configuration choice, I have run databases (Oracle and Informix) over NFS. Make absolutely certain that your are up to date on NFS and Networking patches and that your network is as robust as possible. It could be worse; you might be asked to do this via NFS over the WAN.

You should do as much as you can to make your end of the connection robust (e.g. APA). The main drawback is that the typical things that you can do for redundancy (multiple paths, mirrors) are not possible. You have to trust that the NFS connection works and that nobody decides to reboot the NFS server because their application is hung.

If at all possible, I would lobby for more directly attached storage but it sounds as though the answer is NFS, now what's the question?

[Bill Hassell]

And of course, you are NOT sharing this network link with PCs and other junk, correct? The majority of NFS problems can be traced using a shared LAN with unlimited access (ie, anyone can plugin a hub and hook iPODs, Blackberries, home laptops, etc). If you use a completely isolated network with *NO* router to anywhere, you should be OK. If you need external access to this network for management of the NAS box, use a serial connection to a console multiplexer (you probably have one of these for secure connections to your HP-UX servers and network appliances)

[OldSchool]

i'm not familiar w/ the celera, but i ran ingres for 4 years on a emc ip4700 NAS and it worked fine...

all of test and development were on the ip4700 with the only boot disks in the box. only issue was to make sure that all of the servers sync to same time source. wasn't a database issue, but source code control would get confused if the times were allowed to drift.

[TwoProc]

IF it was on an absolutely private chunk of network that did nothing but this job. I MIGHT consider it (if I had to) if I could have local or SAN attached storage for the redologs. That way, I'd always be able to recover.

However, I (like you) would be trying to avoid that type of deployment. I've done it before (Oracle 7) years ago as a temporary measure, and it worked fine (though slow), but I got off of it as soon as I could.

[Steven E. Protter]

Shalom,

I owed you more explanation, no additional points warranted.

1) We have a private network just for NFS traffic and gigabit hardware end to end for critical servers.
2) The NAS, EMC Celerra uses NFS v3. I have it on good authority that their release with NFS v4 is not stable.

I would agree with the others in saying this setup is far from ideal, but you can make it work.

SEP

[KapilRaj]

I appreciate all the answers. Thank you for spending some time over it.

I am going to do some more tests. i.e. I have to somehow emulate a production workload on this NAS system. My plan is to create some test filesystems and mount them on to several nodes and do contenious read / write operations.

I will come back with more test results. Once again thanks to everyone who participated in this discussion.


Regds,

Kaps

[Alzhy]

There are FULLY SUPPORTED configurations for running Oracle on a NAS (a.k.a. NFS). Most are from NetApp Filers but I beleieve there are also other NAS implementations that are certified (possibly the Celerra is now one). One shop that I've helped last year built their own NAS. It's NAS head is a Solaris 10 machine hooked up via Fibre Channel to a really no well-known Array Vendor. The NAS head hosts about a dozen mix of Linux, Solaris and Windows machines running databases from MySQL to InGress to Oracle 9. All worked and performed well.

The trick to successfully implementing Databases on NFS/NAS are as follows:

1. Use a trunking between your network switch and your NAS head if you cannot have dedicated (i.e. trunked cross-over conections) Gigabit/multi-gigabit connections between your db server and your NAS head.

2. Use fast and zippy CPUs on your NAS heads with lots and lots of memory.

3. Use NFS v3 and UDP protocol (This is true for my Solaris 10 NAS head).

4. If using COmmerical NAS Filers - STRICT adherance to vendor configuration standards MUST BE OBSERVED - specially if the implementation is officially supported by the Database vendor - i.e. Oracle.

That's it. Hope this helps.

[KapilRaj]

I have understood the condition now. I did a test with 8 servers trying to write something at the same and the performance degradation was 5 times. And also we have a single network for everything. I am not ok to keep the oracle database on the NAS now.

Thanks for all your suggessions and your valuable time spent on my request.

Regds,

Kaps