Re: [suse-oracle] SUSE Clustering, Oracle and SAP

I've dealth with many many ocfs2 problems with ocfs2 and SLES10.  Novells
usual response is to upgrade to the latest kernel and patches however that
normally does not resolve the problem at hand.  I've administered both rhel
and sles servers and it seems that rhel's implementation of ocfs2 is far
better than sles and it runs a lot more stable than sles does.  

As far as ocfs2 timeouts what kind of storage are you using ? there is a way
to calculate the timeouts based on the storage your using and what kind of
raid set your set to on the storage side.  

Also watch out for orphaned files on your ocfs2 filesystems which will make
the fs go read-only in sles.

Its too bad that ASM is not an option with SAP using linux since ocfs2 is so
problematic

Justin


-----Original Message-----
From: [email protected]
[mailto:[email protected]] On Behalf Of Miller, Daniel J.
Sent: Thursday, September 09, 2010 9:48 AM
To: [email protected]
Subject: Re: [suse-oracle] SUSE Clustering, Oracle and SAP


For what it's worth.  We've had very good luck with (now CA) XoSoft
WanSyncHA.  It's basically block level storage replication with a watchdog
that fails over the database if the primary goes down.

A couple of caveats are:  

This is not a cluster, so there's no load balancing - the secondary database
isn't even up and it usually takes a couple minutes for the secondary
database to start up before it becomes available (in the event the primary
goes down).   You're down for about 5 minutes (detection, initiate, startup
complete failover) if your database server crashes.

It's asynchronous, so it is possible that some transactions won't have made
it into the queue (This wouldn't work for a bank's on-line system).  We've
not seen anything get lost (we've used XoSoft on a half dozen databases,
some fairly large and busy, for years, but in an asynchronous standby system
transactions can get lost.

Some good news.

Since the oracle instance isn't running on the failover box, the standby box
doesn't need a license (you're limited to, IIRC, 2 or 3 days a month).
Obviously, if your primary server blows up and is down indefinitely, your
license would be transferred to the failover box.

Fail over / Fail back works very, very, well and has been very reliable for
us.  We've been using this for years on windows boxes.  Fail back is often
the sticking point for HA systems.  It's very clean on XoSoft.  They say
they have support for SuSE.  We haven't tested it on SuSE.


==========================
Daniel Miller
Infrastructure Assurance Center
Argonne, Il. 60439


-----Original Message-----
From: [email protected]
[mailto:[email protected]] On Behalf Of Werner Flamme
Sent: Thursday, September 09, 2010 4:26 AM
To: [email protected]
Subject: Re: [suse-oracle] SUSE Clustering, Oracle and SAP

-----BEGIN PGP SIGNED MESSAGE-----
Hash: SHA1

Alexei_Roudnev [02.09.2010 21:27]:
> It don't make much difference - even if someone striped the stuff, it 
> can be installeld back easily.
> 
>>>> Alexei,
>>
>> Hi, sorry to correct you Alexei, since SLES11, the "clustering stuff"
>> was stripped from SLES base and is part of an addon product now...
>> "High Availability Extension". You can download SLES11SP1 and HAE to 
>> use without cost, but as I understand it, you need to pay an extra to 
>> have support for HAE in addition to paying for SLES11 support.
> 
> Support is really a good thing for HA systems. There is a HUGE number 
> of factors you should consider when building them (I spent about 2 
> weeks getting together dell ilmi fenciong, sdb fencing, making it work 
> if server lost all power, making it work on vmware etc. So we have a 
> STANDARD CentOS cluster now, and has standard OpenSuSe cluster. But it 
> really TOOK TIME to R&D them, and a lot of testing.). But it should be 
> more _consulting support_ and not so much _HA components support_.
> 
> Common mistake  - if you add HA stack into the system , it not always 
> increase reliability becasue HA component can fail too. HA systems are 
> oftenn even less reliable then non HA systems (what they always 
> improve is maintaining hardware).

You're so very right Alexej... It's just sad that OCFS2 was moved into the
HA pack too. When connecting my SAP VMs (SLES 11.1) to the RAC (running
Oracle Linux), we had to use either Veritas FS or OCFS2, and thus had to buy
the HA kit. The only thing we use from this is OCFS2.

Now the systems runs for one week without crashing. Up to now, all the
crashes were due to OCFS2 and its timeouts... Still, there is one VM that
stops in the middle of the backup because OCFS2 is remounted read-only
because of an error :-( This causes some grief, and the boxes aren't even on
HA...

Regards,
Werner

-----BEGIN PGP SIGNATURE-----
Version: GnuPG v2.0.15 (GNU/Linux)
Comment: Using GnuPG with SUSE - http://enigmail.mozdev.org/

iEYEARECAAYFAkyIqBsACgkQk33Krq8b42O8EQCfZd2irJ+sUL3fVArOr6TP1d7j
3JQAnj1D406t01+/Ys1f06Qbfa2RxWFP
=gq23
-----END PGP SIGNATURE-----
_______________________________________________
suse-oracle mailing list
[email protected]
http://listx.novell.com/mailman/listinfo/suse-oracle
_______________________________________________
suse-oracle mailing list
[email protected]
http://listx.novell.com/mailman/listinfo/suse-oracle

_______________________________________________
suse-oracle mailing list
[email protected]
http://listx.novell.com/mailman/listinfo/suse-oracle