Agreed - with some serious scripting, use of VIPs, and general clustering experience you can go quite a ways with a commodity NFS setup; the more time and expertise you put into it, the higher the "brain cost."
To be sure, commercial NFS is going to add a lot of features (OpenFiler may indeed replicate some of these - excellent point) but unless there is a larger need for HA NFS in the enterprise, I might want to point my $$ at the skilled OS person vs the commercial filer; I can get a lot of good things out of that person beyond shared application file system / NFS automation. Situations and budgets will of course dictate that decision.
The thing I really like about this setup (regardless of the failover sophistication) is that it employs physically separate servers for failover. NetApp, EMC, etc all have internal redundancy on many models - dual heads, multipathing, etc - but nothing beats having 2 of something for redundancy. It might not always fall under the rubric of high availability, but it's sure nice to know that if one of the units happens to catch on fire or befall some other unspeakable tragedy, the other unit is racked safely racked a couple of cabinets away.
Chuck On Nov 6, 2009, at 10:43 AM, Luis Freitas wrote:
Chuck, I don't agree that this setup can't deliver automated NFS failover.It should be possible to set these boxes with a shared device, DRDB, OCSF2, whatever, and use Oracle CRS to manage the NFS server and virtual IP.The Linux NFS server already has some provisions to allow for a high availability deployment. There are some parts missing, like proper nfs locks replay, but it can be made to work as an application file system, where these bits are not critical. Not sure on how this goes with other vendors, as they would expect you to purchase their clusterware.I am playing with a configuration like this here, with the added complication of mixing Solaris clients with the Linux NFS server, over a OCFS2 filesystem. But we will purchase a NAS device next year so we may not need to go into production with this setup. To ensure high availability we mounted the filesystem using one of the RAC VIPs already setup on the server, since the NFS server is also a RAC database server, and added some configuration parameters on the NFS server to allow for a transparent reconnection on the client side.But a NetApp filer provides a lot of functionality, like automated snapshots, replication across remote or DR sites, etc.Also we should not consider only the hardware cost, but the overal cost. Implementing a solution like this with linux or Unix will need work from a specialized O/S consultant, and also the post- implementation suport cost, for which the filer could have better management tools. Depending on the size of the deployment the suport costs offset the initial implementation cost.Btw, you could install OpenFiler on those two boxes. I don't know if the High Availability bits are on the community edition.Best Regards, Luis Freitas --- On Fri, 11/6/09, Chuck Edwards <chuck@xxxxxxxxxxxxx> wrote:From: Chuck Edwards <chuck@xxxxxxxxxxxxx> Subject: Re: Shared APPL_TOP filesystem question To: ora-apps-dba@xxxxxxxxxxxxx Date: Friday, November 6, 2009, 4:03 PM From a performance requirement perspective, commercial NFS is overkill for shared application file systems. Standard Linux/UNIX NFS will do just fine and is quite cheap. For failover, you can simply set up two servers and have them rsync changes from one to another. If you experience a failure, simply mount the file system from the failover on your application servers and off you go. It's not instantaneous, to be sure, but it provides a couple of advantages: 1. Cost is very low. A pair of ~$900 white boxes with mirrored disk and 2 - 4GB of memory will do nicely. 2. Since you have two separate servers, you can place them in physically separate cabinets, an advantage that even expensive, internally-redundant NFS filers don't enjoy. Again, if instant, auto-magical failover for shared application file systems is an absolute requirement, this solution cannot deliver; however, performance would be just fine. ( If you take minimal steps to pre-configure mount points and /etc/fstab, then script up the mounting and dismounting, failover can be quite fast, though. ) I suspect that with some of the new ASM capabilities in the 11.2 database release, we're going to see certification for shared application file systems in ASM, which will change this discussion substantially. Chuck Edwards chuck@xxxxxxxxxxxxx On Nov 6, 2009, at 9:48 AM, Luis Freitas wrote:Hmmm, NFS thirdy party vendors are $$$, butoperating system clusterware to manage NFS high availability can also be $$$.Sun cluster suite, HP cluster, orVeritas is not exactly cheap. I don't know how the clusterware licensing goes for AIX. For Redhat, RHCS used to be licensed separatelly but on the latest release they included it on the operating system license.Of course you can use Oracle CRS, but ifthe NFS servers doesn't have any other oracle products, you need to pay for a license too, and it won't be integrated with the O/S NFS server, so you can't expect any support from your OS vendor with the failover procedures that need to be implemented on CRS.Best Regards, Luis Freitas