Haven't had a chance to try this for real, but had the idea a few nights ago.
Given that the oracle client runs very nicely off nfs, and that shared oracle homes for the server are now supported as of version 10
If we have a VE connected to shared storage (SAN would be nice, though oracle does claim to support NFS) running RAC / Grid, it should be possible to simply add another node to the VE, run your favorite RAC management tool or OEM to add a node to a db cluster, edit your tnsnames or sqlnet file on the shared client
Volia. Instant oracle scalability. No more digging out ancient glibc builds, kernel tuning, etc. The shared image has the OS as well as the app
yes, you are right that this should work.
It is exactly the way how shared filesystem-images in openQRM are supposed to be used.
There is even a private-directory mechanism so you can configure your clients in a static way while using them dynamically.
For a SAN-based configuration you could use the built-in Iscsi-support to boot your nodes directly from an Iscsi-target. What you will need in this case is a robust cluster-filesystem to support mulitple-parallel-client access.
I made some experiences with the combination of ocfs2 and Iscsi on openQRM and it worked very well in my tests. If you plan to set this up i would be more than happy to assist.
One thing I see is that there isn't any mention of the $60k-80k per processor list price one would have to shell out for an Oracle license. So wouldn't it be better to use somethng like PGSql for something like this?
Keeping it real,
You get what you pay for. Postgres is nice and all but it doesn't, afaik, support active / active clustering
I can run a query against a specific instance of RAC, pull the power out of that node, and watch the query continue uninterrupted on another instance. Don't recall postgres doing that
I've got exactly the same idea for the company I work for.
We are currently redesigning our IT landscape.
And one of the things that we're going to do is have an Oracle RAC grid with several different nodes (e-business apps) (DB).
with OpenQRM I envision it to be possibele to do flexable provisioning of nodes in the grid exactly there where you need the performance.
just boot up an extra node. and bring down a node where you don't need it to go into standby mode.
However doing this with Suse Linux on various boxes brings me a big challenge.
the whole package is kind of RH centric (snif)
we don't have redhat here so all sorts of issues to overcome..
hope to get over this really soon (few weeks)
but so far I only was able to install and setup the QRM server and create images (kernel and filesystem) for test purposes.
PXE booting fails at the point where the / filesystem needs to be mounted over nfs (for the time being).
Production grid will probably be infiniband.
I'm interested in why the network booting fails.
What version of openQRM are you using?
Do idle nodes come up ok?
What NFS server are you using?
What error message do you get on the console when the booting fails?
I'm not sure your correct on the support active / active clustering point. However, I'm also no expert. Just a fan of cheap dbs. ;-]
Off to do some research,
The issues around support active and active clustering could probably be cleared by decibel if he's out there today ;)
Decibel? Can you help us with this since you're the
"official" postgres master for OpenQRM?
Thanks in advance,
i posted a reply to this and your other thread into the suggestion-box forum.
thanks and regards,