HP-UX with LVM and VxVM switch migration

Question

Hi guys, I have a HP-UX 11.23 server with LVM VGs and VxVM 4.1 DGs. PowerPath 5.0.0 fully licensed. It has access to more than 130 luns on a DMX3000 through 2 switches. MDS9509 and McData ED-140M. I need to migrate both the HBAs of the server and the FAs it sees to new switches (MDS). Since the new switches cannot have the same DomainID of the old switches (already configured), the FCID's of the server and the DMX will change, causing the /dev/ctd's to change as well. I know I can move one path, reconfigure the LVM with the new ctds by doing 'vgextend' and 'vgreduce'. And do the same later for the other path. However, I do not know how will the VxVM behave. Since I only see one address (ctd), how can do the re-configuration of the DGs? Will the PowerPath automatically protect me during this? And how will the server behave after a reboot? I want to do this migration as smooth as possible and without any downtime. So please help me guys Thanks in advance, Maxim

SKT2 · Answer

is the node part of a cluster?

MaximUnited · Answer

Yes. HP service guard. I think A.11.17I have to move both nodes.Does it make a difference?

SKT2 · Answer

i am not sure about the VxVM. But for the LVM/VG part the the ctd number of the cluster lock pv would change if that disk is also part of migration. So you have to update the cluster configuration which is normally a offline(cluster) activity with MCSG 11.17 too.

MaximUnited · Answer

These are bad news... Can someone confirm that?And I need an answer about the VxVM please...Thanks

xe2sdc · Answer

If your MC/SG doesn't use a Quorum Server, changing Quorum requires an offline of the whole cluster.
I don't have details regarding VxVM ... but as long as you have to change the cluster lock device, you need to shutdown the cluster.

dynamox · Answer

Santhosh,i know that if you change cluster lock disk, you have to re-compile the cluster, but i remember you mentioned one time that you can change cluster lock pv without recompiling the cluster (can't find that thread). It's still an off-line process but at least no messy cluster recompiling ?

SKT2 · Answer

the work around we used is keeping /recreating(mknod) the existing Ctd device with the same minor number of the new device/path. This way if an attempt is made by the cluster to the old CTD device that will go/reach to the new path.

dynamox · Answer

cool ..thank you for reminding me.

MaximUnited · Answer

santhosh,How do I do that? I'm not that familiar with all system tricks...Guys, I have to do the migrations for Solaris hosts too.I'm pretty sure all filesystems are pointing to the psedo devices (emcpower1c, etc).So that migration will go smoothly? I need to keep something in mind?THX.

Host Systems

Was this post helpful?