Originally posted by: SystemAdmin

We recently had a system with +5.5TB data (Mirrored) changed to MPIO and Enh. Concurrent VG running SAP.
.
Very soon we had system hangs with :
LVDD QUORUM LOST, VOLUME GROUP CLOSING.
LVM_GS_RLEAVE
.
The fix suggested that MWC be turned off to make the configuration legal. Sample dialog in the 5.4.1 HACMP documentation.
We used MWC/Passive.
Oracle documentation does not like this idea.
.
IBM documentation suggests with MWC off coming back up, mirror the VG first and that let the application decide if the data is OK.
syncvg -f -l LTVname
This could take a VERY long time and you could mirror good data over bad data?
.
Question is:
Is this a good idea to use MPIO?
Or are there other measures I am unaware of that makes it not so bad?
Are there any documentation discussing this MWC to be off with MPIO and ELVs for this makes a fall over to be potentially very slow?
.
We are a bit uneasy with this.
Jan
#AIX-Forum

Originally posted by: alethad

I ran into the LVM_GS_RLEAVE with one of our systems a few months ago. It was to my understanding that is is an AIX OS issue although it affects HACMP systems. The problem is with the gsclvmd daemon which isn't exclusive to HACMP.
I subscribed to recieve an email when the fix was available but have not heard anything yet.
Is the fix included with the HACMP 5.4.1?

We switched the mpio last fall and have not had any issues with it except for this one. We were using sddpcm.

We are a HACMP/Oracle shop also. I'm not sure what you mean by MWC. What is that?
#AIX-Forum

Originally posted by: alethad

Jan,
Our Oracle systems are not set up with raw partitions so I don't have to worry about MWC. We mirror across 2 IBM sharks using AIX mirroring not PPRC.
Our clusters are set up Enh. Con. since that is required by HA but only 1 of the nodes uses a VG at any given time. We don't allow actual concurrent access to our disks.

Good luck.
#AIX-Forum

Originally posted by: SystemAdmin

Hi

Here a link to MWC.
http://publib.boulder.ibm.com/infocenter/systems/index.jsp?topic=/com.ibm.aix.baseadmn/doc/baseadmndita/mirwritelogvol.htm

#AIX-Forum

Originally posted by: jvk

I think that when one uses LVM mirroring across two storage boxes then it's always a good idea to enable MWC. MPIO has nothing with this..
If you disable MWC you could have data consistency issues if you crash and have to recover. It will reduce the number of writes by 50%, but after a crash a forced syncvg should be run. This can take a long time..

However if you use HACMP in enhanced-concurrent mode then (i think) enhanced-concurrent mode does not support MWC...
For sure Mirror Write Consistency (MWC) (and Bad Block Relocation too) are not supported in a concurrent setup with multiple active nodes accessing a disk at the same time. See 'man chlv' for this.
Disabling MWC is the right thing to do to avoid potential race condition in meta data operation. After a crash a forced syncvg should be run.

Hmm, what about case when only one node is using the VG and you use the concurrent option just for the fast takeover, in this case maybe you can switch the MWC on..
Maybe this doc can help you too
http://technology.amis.nl/blog/?p=1156
#AIX-Forum

Originally posted by: SystemAdmin

jkv

This is the question to be answered as you said.
"what about in the case when only one node is using the VG and you use the concurrent option just for the fast takeover, in this case maybe you can switch the MWC on."
We do not use Oracle RAC.

I agree that MPIO does not play a role here except that, that forces us to use Concurrent Capable VGs. This made IBM say we must disable MWC. That leads to a long time in running a syncvg before activating the filesystems after a crash.

Could we just not make use of Concurrent Capable VGs with MPIO.
One might then, to save time in a fallover, not use HACMP and fallover manually?
Defeats the object a bit.

Does gsclvmd on the standby node write to the VG varied on "concurrent/passive"?

As usual, more questions!
5300-08-02-0822
HA 5.4.1 SP02
#AIX-Forum

Originally posted by: jvk

Yes, right info is that MWC on enhanced concurrent mode VGs is not supported if and only if the VGs will be used concurrently (e.g. GPFS or Oracle RAC). Since your HACMP setup only has non-concurrent resource groups, you should be okay when turning MWC on.
#AIX-Forum

Originally posted by: SystemAdmin

Some more Info.

US Patent 7096333 - Limited concurrent host access in a logical volume management data storage environment

http://www.patentstorm.us/patents/7096333/description.html
#AIX-Forum

Originally posted by: SystemAdmin

We have a valid configuration.
MWC on with Enh.Conc.VG in active/passive mode.
IBM still investigating our problem.
#AIX-Forum