mirror MC/SG problem

[Javier Ortiz Guajardo]

I have 2 vg??s defined in a MC/Service Guard (vg11 and vg12) its Logical Volumes are mirrored. We define this vg??s adding the primary disk first (c5txdy Disk with best performance than c20) and then the Mirrored disk (c20txdy)
With comand lvdisplay -v /dev/vg11/lvol1 you see this
---- Logical Extents ------
LE PV1 PE1 Status 1 PV2 PE2 Status 2
0000 /dev/dsk/c5t0d0 0000 current /dev/dsk/c20t0d0 0000 current

After a reboot, we start MC/SG package and it activates the vg11 and vg12, but i see know that the c20 is taking the place of c5.
---- Logical Extents ------
LE PV1 PE1 Status 1 PV2 PE2 Status 2
0000 /dev/dsk/c20t0d0 0000 current /dev/dsk/c5t0d0 0000 current

Please, i want you to help me to know why the system changes the disks?
it means that i??m using the disc with less performance?
Is a Mirror Problem?
thanks for your help.

i attach the lvmtab file and the vg11.conf
my email is jeortizg@hotmail.com

[Tim D Fulford]

It depends on how your /etc/lvmtab is laid out. The first disk it comes acress is the primary etc. (strings /etc/lvmtab will show you this)

You may have swapped it 'round by doing pvchange -s y /dev/dsk/....
This is temporary & the next re-boot will plip it back.
OR
when you add a logical volume to one system you do a vgexport -m /tmp/vgxx.map -s -p -v vgxx & then a vgimport -m /tmp/vgxx.map -s -v vgxx. Although this usually imports it in lowest number first.

If you want to make it more permanent...
dont do vgexport & vgimport with the -s flag. so do
vgimport -m /tmp/vgxx.map -v vgxx /dev/dsk/... /dev/dsk/...

& OR keep a copy of a "good" /etc/lvmtab.good & copy it back onto the /etc/lvmtab. this option is probably far from supported!!!

Regards

Tim

[Dietmar Konermann]

After initial activation of a VG the PVs with the smallest key should be displayed on the left side. You may check this using the -k option with lvdisplay -v. The keys are assigned in increasing order when the PVs are added to the VG, I assume. So it looks like c5tod0 was not the 1st disk of this VG.

BTW, don't expect too much performance penalty... writes go always to both disks and reads go to the disk with the sortest queue.

Best regards...
Dietmar.

[Javier Ortiz Guajardo]

Can you tell me more about the smallest key
si it configurable? is HW definition ?

thanks.

[Suresh Patoria]

Hi,

If you see the /etc/lvmtab file then you will not find the entry of C20t0d0

So according to me rename the /etc/lvmtab file and create the new file with

vgscan -a -v

This command will scan all the logical volume and create the /etc/lvmtab file

After this You check the lvdisplay -v /dev/vg11/lvol1 command

[AlienRoadShow]

It seems possible that your cables are reversed. This may or may not be the source of your problem.

What model servers please?

HW addresses are determined by the slot position in the system board. While disk controller target numbers are based upon the HW address.

For example here are a couple of L class / rp54## server slots:

9
Turbo
0/6/0
Turbo
0/6/0
Turbo
0/9/0

8
Turbo
0/2/0
Turbo
0/2/0
Turbo
0/3/0

See this link:

http://www.docs.hp.com/hpux/onlinedocs/A5191-96022/A5191-96022.html

[Stuart Abramson_2]

I don't understand why one path has better or worse performance than the other. Maybe you assigned PVs unequally to paths and have more PVs on one path then the other.

But, you can change primary path with command:

2. pvchange -s /dev/dsk/"old alternate"

This will make the old alternate into the new primary. However,
it is only temporary, and on the next reboot or package halt/restart,
it will revert to the original configuration.

[Tim D Fulford]

Stuart

- You alternate paths to spread the IO load over as many interfaces as possible. Even when mirroring one "disk" will be more active than the other. (write does equal IO, read IO only required from one disk)

- pvchange ... Will work, until the next boot when it will revert back to its original state.

Just my 0.02??? worth

Tim