ASM Raw versus LVM/VXvM RAW on HP-UX 11.11and Oracle 10G? XP/HDS-USP Storage..

[Alzhy]

Greetings!

Our client seem to be only partial to having RAW storage of our MEGA databases on RAW that is if we are to use ASM with it.

I am still of the opinion we will be better off just using manual RAW storage configuration either using LVM, VxVM or SAN pre-sized disks/LUNs..

Can ASM already be trusted better than the ertshwhile ODM (Oracle Disk Manager)? We do not have RAC environments.

Any inputs / opinions from those that actually have used or are using ASM and or RAW devices under Oracle 9/10/11G are welcome... Hearsay opinions are welcome too...

Thanks.

[Hein van den Heuvel · Accepted solution]

Hi Nelson,

(HPTF in two weeks?)

I just finished an Oracle 11G, RAC, on ASM (bigfiles) install on Weendoze.
Went pretty smooth... (the 5th time. :-).
I liked the ASM part but my installation was not big enough to really exploit it just yet.

Still, even for this simple setup the reduced management was already visible.
And then there is the promis of load balancing and ease of storage expansion/movement. Trule Promissing!
Not sure how it holds up 'under pressure' in complex multi-db setups. It feels a little like putting all your eggs in one basket.

Anyway, fwiw, my first peek looked good.
I was a little worried about yet another wart on my DB to worry about, but it is just a simple little (rac) database on the side and it speaks SQLplus if you like, so no big deal.

2c,

Hein.

[Duncan Edmonstone]

Well all opinion is subjective, but I generally think that Oracle would like everyone to run their databases on Oracle Unbreakable Linux with RAC and ASM, and their technology reflects that... On Linux ASM is pretty good, on other platforms... well take a look in Metalink and you'll find *plenty* of issues.

My 2 big beefs with ASM are:

i) No multipathing - so you still have to provide that yourself - not an issue on 11iv3 I guess, but on v1/v2 you need to use the array vendors sw (I guess HDLM in your case Nelson)

ii) Now all database storage management is under the control of Oracle - including backup (and snapshots if your snapshot technnology doesn't support a "consistent split" mechanism). Oracle seem to exist in this dream where "all the world's a database" - it isn't and there are other things I want to store on my system - why do I want more than one volume management mechanism to deal with, and since when were DBAs storage management gurus? (many of the DBAs I've dealt with over the years are still stuck in the old OFA "one spindle per filesystem" paradigm.)

Performance-wise, I've seen differing stories - I think in reality with the correct tuning the differences are <10% so who cares?

HTH

Duncan

[Alzhy]

Thanks Hein and Duncan,

And see you MigZ at HPTF 2008...

The reason really why we're talking about ASM (and probably ODM in its absence) is because the client would only allow us to go RAW with it - period.

IMO, we will be better off with just using LVM or VxVM RAW as we already know how to make optimal storage layouts and so we as storage and system admins can have better control and idea of what's going on.

You mentioned Multi-pating - that will indeed be an issue as we're only licensed for VxVM with HDLM (or AutoPath) needing to take the DMP/LB chores from VxVM. We're stuck with HP-UX 11.11 BTW...

ANd I thought ASM is pretty much conducive only for RAC implementations. Ours is not a RAC implementation.

Hope others that are using ASM in a non-RAC scheme chime in.

[Duncan Edmonstone]

One point about ASM in non-RAC environments - it doesn't work with Serviceguard, so if you want to do HA with an ASM solution you are pretty much pushed down the RAC route...

HTH

Duncan

[Alzhy]

Update:

We're ASM bound my friends. After trying it on Linux, it's all RAW so anything sashimi for me is good!

Yep, ASM indeed does not support Multi-Pathing by itself. It needs to have raw disks (or devices) that are already path protected or laod balanced. And since our environments are heavy on VxVM, we just can't drop VxVM's DMP.

It turns out, Oracle and Symantec came out with a recipe to allow ASM to exploit VxVM DMP configurations. Best practice is to create volumes wholly on each member disk in a diskgroup. The resultant volume will now be representative of a "disk" but load balanced/protected.

In Oracle ASM gyrations, you simply tell the instance to set its "asm_diskstring" to point to your Diskgroup raw directory.

For example if youre VxVM Diskgroup is named asm_dg and your disks were converted into volumes named asmdisk001 to asmdisk999, then :

asm_diskstring='/dev/vx/rdsk/asm_dg/asm*'

We're benchmarking and comparing ASM raw with traditional VxVM fronted RAW and LVM fronted raw if there's any drag from having VxVM as the path and volume protection under ASM.

The recipe from Symantec and Oracle (for those who'll need it):

http://eval.symantec.com/mktginfo/enterprise/white_papers/b-whitepaper_integrating_oracle_automatic_01-2008.en-us.pdf

[TwoProc]

Maaaaannnn,

I would be scared to death to rely totally on Oracle for all I/O. Just think of when you go to do a recovery, etc, and you find out you didn't have the last patch bundle in place to ensure a good backup... or recovery...

I know you can't do it at this site, for me, if I had a place that wanted raw, I'd just use lvm if I could.

HP is well known for robust servers and file systems for eons, and Oracle is known for ... its customer support...

SO ... let's talk about being in a critical state and needing emergency support... I'm sure they'll get that ONE late night super duper DBA left on staff since outsourcing everything overseas to help you, if he or she not is already busy on another call... But, they'll respond to your SEV1 within two hours, they promise.

Unfortunately, their response will be:

Have you run and posted the results of the db_info_gather_stuff_about_your_system_of_the_week
java thingy?

And then, in 5 hours after you've run it:
"I see that section xxx didn't run correctly.
Could you patch up the version you have to the latest version on our site?"

And you post these new stuff, which looks exactly like the prior results.

And then, that evening, a decisive response!

"I think your ASM system is corrupt. Try laying it all out again and recovering, then get back to us and let us know how that works for you. In the meantime, we're going to move this to a Sev2 until you get back to us..."

Bwaaah Bwaaaahhh Bwaaaaaaahhhhhhhhhh.

It's OK though, 'cause you've had a day to put out your resume on youbetchajob.com and you've even had a couple of hits from some recruiter-bots - or were they just spamming you to sell you the secret to adding water to your gasoline to improve your MPG?

[Alzhy]

TwoProc,

Thanks man and I hear ya!

I wanted simple RAw with us admins and dba's still knowing what's really going on as far as I/O.

But you see.. this mode of RAW Oracle storage is the only one allowed by our "brave" and "sophisticated" client.