RHEL versus OL

[lewonkas]

Greeting UNYOUG-ers, not really clear how used this board is, but here goes anyways.
After a lot of years building, using and generally "speaking well of" Oracle/Oracle RAC on Windows, we're parting ways (with the Windows part that is; )

We're planning/budgeting initial projects to replace all our Oracle database environments on Windows OSes and move to Linux x86-64.
I'd love to hear some of your distribution preferences and why.

We have next-to-no Linux in our shop today -- a couple hosts running MRTG & RRD. We have no predispostion any particular distribution, but we'd realisticly be choosing 'tween RHEL and OL as the primary contenders. We don't do findge-anything, which is what Oracle on Windows has been becoming, but i digress....

One important point is that this is seen as purely for our Oracle DB server environment. No plans for servers beyond that.

thanks - Steve

[Jer]

Honestly I would keep it with OL. There's something to be said about doing all of the management with one support shop. If you run into an issue with RHEL and you are putting in a ticket to Oracle, you might get caught in "not my problem land" with RedHat support and that's never fun. Oracle has always been good about supporting their stuff on RedHat (I've never had issues) but I will tell you that Oracle's response time is better than RedHat's, and that can make a difference when something critical isn't working.

Congrats on making the switch to Linux for your Oracle servers. I was able to do some filesystem tuning with my Redhat boxes and get a 30% performance edge over Windows in some of my data warehousing situations.

~Jer

[lewonkas]

Thanks Jer, that makes complete sense. We today have no relationship with Red Hat, the couple RHELs we run are unsupported.
There would have to be a compelling reason to Introduce another vendor into the mix. We're looking to started with OL 5.7 or 5.8. I've been wading through the tangle of server hardware OS certifications. Not quite ready for the jump to 6.x as i'll still need to support some 10.2 rel databases. thanks

[Jer]

6.x is a little messy anyway, even with 11gR1.
The great thing about Linux is that it will run on almost anything!
The most important hardware to certify is your HBU and RAID cards; I've seen IO problems when a generic driver is used with a high-end IO device.

~Jer

[lewonkas]

Thanks for the reposonse, been busy and just getting back to my Oracle on Linux project.
Could you elborate on the 6.x issue? what have you found?
as far as certifications go, we were going to try to keep within the certified hardware choices. We are a Dell shop (largely) and are researching OL versions and server models. We have an upcoming meeting with our Dell contacts about upcoming certifications of there new "12g" servers, specifically the R720. These are small boxes - 2 socket (for a 2 node RAC) . We're starting with our Oracle SE implementation.

I have not seen certification specific to HBAs -- anything you can point me too? Our OL environment would be on dual attached to SAN, backend array would be IBM XIV Gen3. We've just selected this new array, so I have yet to dig into the host attachement kits for this.
Thanks!

[Jer]

I just had a little bit of difficulty getting 11.1.x running on 6.2 because the libraries and packages were *so* different (as they should be, 11.1.x is nearly 5 years old now). I'm sure if you run 6.2 then Oracle 11.2 (or 12.1) works great.

Regarding hardware certifications, you can find most of that info here: https://hardware.redhat.com/ (Since RHEL and OEL share binaries, the same list will apply to OEL).

Since Dell actually sells RedHat, most of the configurations that you can get RH with a Dell server would be fine. Just make sure you do some Google sleuthing on the HBA and customizations they offer to make sure that there aren't known issues with those components.

As a side note on storage, I'd like to point you to a few articles:
http://docs.oracle.com/cd/E11882_01/server.112/e18951/asmprepare.htm
http://docs.oracle.com/cd/E11882_01/install.112/e17212/storage.htm

I don't know whether the XIV3 does lazy commits to disk from the SSD, but before the SSD arrays it was important to keep your logs (redo/undo) on a LUN that didn't depend on parity calculations. I haven't used any of the newer arrays to test them, but it might not be so much of an issue anymore. I would say test it if your storage admin is pressuring you to do RAID5/6.

[lewonkas]

Thanks for the storage doc link, i'll take a look. We'll have some help from IBM on the specifics of the configuration -- XIV and Oracle seems to be a farily popular configuration, with a number of IBM tech docs on it. Also had a number of good reports from customers running Oracle on it.
We are not (at least initially) deploying the SSD cache on the XIV. XIV also does it own brank of data protection -- not RAID 5/6.
I have yet to see any specific XIV recommendation fo the handling redo, but i'll ask the question. Tho there are best practices about the number of LUNs per ASM diskgroup and AU size (fewer LUNS and large AU sizes). We'll stick to what's recommened, but In gerneral we are looking for the simplest pool/LUN configuration possible.
First array gets here early July!
thanks Jer