Re: [Ocfs2-users] OCFS2 vs Oracle ASM CFS

From: Jeremy Schneider <jeremy.schneider_at_ardentperf.com>
Date: Thu, 03 Sep 2009 14:48:05 -0500
Message-ID: <4AA01D75.7020008_at_ardentperf.com>



Thanks Sunil (and Joel) -

Sorry, I typed that email a little too fast - I meant to say ZFS (from Sun), not XFS. I'll have to read up on the REFLINKS -- apologies that I hadn't caught that before; it sounds quite good!

While I disagree quite strongly with the sentiment, it's amazing how many people I still come across who view OCFS2 as a very "immature" filesystem. Very recently, some folks I know who are running hundreds of RAC on Solaris deployments (with shared oracle homes on a cluster filesystem) came up with their Linux RAC plans. They could potentiall roll out a few hundred clusters of this configuration too... and they chose to complicate their deployment by switching OCFS2 mounts to non-cluster local filesystem whenever possible -- in large part because they didn't view OCFS2 as "reliable" enough. (I really tried as hard as I could to convince them otherwise... IMHO it was a poor decision.) With the announcement of ACFS, I imagine that the FUD will persist... so it's good to hear that you guys are regularly stress testing your OCFS2 releases with database workloads. Sometime I'd be interested to hear more about what kind of stress testing you do, especially for non-datafiles (oracle homes and other general purpose uses).

Also, on a side note, I was recently at a site where they create tempfiles as sparse on solaris; not sure if that might be an additional [1.4?] feature of OCFS2 that would be useful for databases.

And thanks again for the response... I guess the hazard of a public list is that you get the occasional dumb email, and after re-reading it, I think that my earlier email sounded a little more obnoxious than I meant it to... I'm glad you put up with me and gave such great feedback. :)

-Jeremy

Sunil Mushran wrote:
> Jeremy Schneider wrote:
>> Some quick questions for the ocfs2-users list, now that 11.2 is
>> public. (Now that it's public knowledge that Oracle developed a whole
>> new
>> [propriatery] cluster filesystem from the ground up for RAC.)
>>
>> 1) Seems that the 11.2 docs encourage using ACFS rather than OCFS2 for
>> oracle binaries. Would it be safe to say that OCFS2 isn't strategic at
>> all anymore for Oracle database files? I've also noticed that OCFS2 is
>> the underlying filesystem in Oracle VM. Is this a shift in strategy for
>> OCFS2? As a development team, are you guys increasing your focus on
>> Oracle VM? I wouldn't expect that Oracle would continue to invest in
>> OCFS2 development only for the benefit of SLES...
>
> OCFS2 was not developed for any one use case. It was always meant to
> be for
> general purpose usage. While most of our early users were using it
> primarily
> with the database, we have come a long way since then.
>
> There is no shift in strategy. OCFS2 is still stress tested with the
> database
> by multiple groups. As far as development effort goes, we took care of
> db usage
> in our first release. In 1.0 we had support for directio, asyncio,
> extents and
> large files. And the performance was in line with raw. Frankly, the
> database
> does not expect much more than that from the file system.
>
> In 1.4, we added unwritten extents that could one day be used by the
> database.
> But mostly the new features were targeted towards general purpose usage.
>
> The story will not change in 1.6.
> http://oss.oracle.com/osswiki/OCFS2/Roadmap
>
> Yes, OracleVM is an important user of OCFS2. So we are obviously
> focused on it.
> But not at the expense of other users.
>
> Oracle is not focused on OCFS2 for SLES. Novell takes care of that.
> Oracle is
> focused on OCFS2 for EL.
>
>> 2) I'm curious if anyone has tried running OCFS2 on DVM? Would there be
>> any point to using OCFS2 rather than ACFS?
>
> I believe OCFS2 has been tested with ADVM. It works with one big caveat.
> In OCFS2, we heartbeat in the fs itself. That is a problem if ADVM
> were to
> freeze the ios temporarily for say resilvering. We intend to fix this
> issue by
> dusting off the global heartbeat patches that allows users to specify
> a different
> heartbeat device. Hopefully in 1.6.
>
>> 3) ACFS shapshots -- or are those a feature of DVM? Seems like a great
>> feature - isn't xfs the only other currently released filesystem to
>> support this? Any plans to add snapshots to the OCFS2 roadmap?
>
> No, XFS does not support snapshots. The only mainstream Linux fs to
> support
> snapshots is BTRFS. (unlimited writable snapshots). Still not PROD
> though.
> http://oss.oracle.com/projects/btrfs/
>
> We have no current plans on supporting traditional volume based
> snapshots.
> What we are working on instead is REFLINKs, which is inode-based writable
> snapshots. I talked about it in the email announcing the 1.4.2 release.
> http://oss.oracle.com/pipermail/ocfs2-announce/2009-June/000028.html
>
>> This is why I've always loved OCFS2 so much to start with -- there is no
>> "asm-users" or "asm-devel" mailing list. :)
>>
>> -Jeremy
>>
>>
>> http://download.oracle.com/docs/cd/E11882_01/server.112/e10500/asmfilesystem.htm
>>
>

-- 
Jeremy Schneider
Chicago, IL
http://www.ardentperf.com

--
http://www.freelists.org/webpage/oracle-l
Received on Thu Sep 03 2009 - 14:48:05 CDT

Original text of this message